Data augmentation is a widely used technique to address the problem of text classification when there is a limited amount of training data. Recent work often tackles this problem using large language models (LLMs) like GPT3 that can generate new examples given already available ones. In this work, we propose a method to generate more helpful augmented data by utilizing the LLM’s abilities to follow instructions and perform few-shot classifications. Our specific PromptMix method consists of two steps: 1) generate challenging text augmentations near class boundaries; however, generating borderline examples increases the risk of false positives in the dataset, so we 2) relabel the text augmentations using a prompting-based LLM classifier to enhance the correctness of labels in the generated data. We evaluate the proposed method in challenging 2-shot and zero-shot settings on four text classification datasets: Banking77, TREC6, Subjectivity (SUBJ), and Twitter Complaints. Our experiments show that generating and, crucially, relabeling borderline examples facilitates the transfer of knowledge of a massive LLM like GPT3.5-turbo into smaller and cheaper classifiers like DistilBERT-base and BERT-base. Furthermore, 2-shot PromptMix outperforms multiple 5-shot data augmentation methods on the four datasets. Our code is available at https://github.com/ServiceNow/PromptMix-EMNLP-2023.

Text style transfer is an important task in natural language generation, which aims to control certain attributes in the generated text, such as politeness, emotion, humor, and many others. It has a long history in the field of natural language processing, and recently has re-gained significant attention thanks to the promising performance brought by deep neural models. In this article, we present a systematic survey of the research on neural text style transfer, spanning over 100 representative articles since the first neural text style transfer work in 2017. We discuss the task formulation, existing datasets and subtasks, evaluation, as well as the rich methodologies in the presence of parallel and non-parallel data. We also provide discussions on a variety of important topics regarding the future development of this task.1

Learning disentangled representations of realworld data is a challenging open problem. Most previous methods have focused on either supervised approaches which use attribute labels or unsupervised approaches that manipulate the factorization in the latent space of models such as the variational autoencoder (VAE) by training with task-specific losses. In this work, we propose polarized-VAE, an approach that disentangles select attributes in the latent space based on proximity measures reflecting the similarity between data points with respect to these attributes. We apply our method to disentangle the semantics and syntax of sentences and carry out transfer experiments. Polarized-VAE outperforms the VAE baseline and is competitive with state-of-the-art approaches, while being more a general framework that is applicable to other attribute disentanglement tasks.

Effective fusion of data from multiple modalities, such as video, speech, and text, is challenging due to the heterogeneous nature of multimodal data. In this paper, we propose adaptive fusion techniques that aim to model context from different modalities effectively. Instead of defining a deterministic fusion operation, such as concatenation, for the network, we let the network decide “how” to combine a given set of multimodal features more effectively. We propose two networks: 1) Auto-Fusion, which learns to compress information from different modalities while preserving the context, and 2) GAN-Fusion, which regularizes the learned latent space given context from complementing modalities. A quantitative evaluation on the tasks of multimodal machine translation and emotion recognition suggests that our lightweight, adaptive networks can better model context from other modalities than existing methods, many of which employ massive transformer-based networks.

Generating relevant responses in a dialog is challenging, and requires not only proper modeling of context in the conversation, but also being able to generate fluent sentences during inference. In this paper, we propose a two-step framework based on generative adversarial nets for generating conditioned responses. Our model first learns a meaningful representation of sentences by autoencoding, and then learns to map an input query to the response representation, which is in turn decoded as a response sentence. Both quantitative and qualitative evaluations show that our model generates more fluent, relevant, and diverse responses than existing state-of-the-art methods.

Automatic sentence summarization produces a shorter version of a sentence, while preserving its most important information. A good summary is characterized by language fluency and high information overlap with the source sentence. We model these two aspects in an unsupervised objective function, consisting of language modeling and semantic similarity metrics. We search for a high-scoring summary by discrete optimization. Our proposed method achieves a new state-of-the art for unsupervised sentence summarization according to ROUGE scores. Additionally, we demonstrate that the commonly reported ROUGE F1 metric is sensitive to summary length. Since this is unwillingly exploited in recent work, we emphasize that future evaluation should explicitly group summarization systems by output length brackets.

We present a novel iterative, edit-based approach to unsupervised sentence simplification. Our model is guided by a scoring function involving fluency, simplicity, and meaning preservation. Then, we iteratively perform word and phrase-level edits on the complex sentence. Compared with previous approaches, our model does not require a parallel training set, but is more controllable and interpretable. Experiments on Newsela and WikiLarge datasets show that our approach is nearly as effective as state-of-the-art supervised approaches.

Text generation has played an important role in various applications of natural language processing (NLP), and kn recent studies, researchers are paying increasing attention to modeling and manipulating the style of the generation text, which we call stylized text generation. In this tutorial, we will provide a comprehensive literature review in this direction. We start from the definition of style and different settings of stylized text generation, illustrated with various applications. Then, we present different settings of stylized generation, such as style-conditioned generation, style-transfer generation, and style-adversarial generation. In each setting, we delve deep into machine learning methods, including embedding learning techniques to represent style, adversarial learning, and reinforcement learning with cycle consistency to match content but to distinguish different styles. We also introduce current approaches to evaluating stylized text generation systems. We conclude our tutorial by presenting the challenges of stylized text generation and discussing future directions, such as small-data training, non-categorical style modeling, and a generalized scope of style transfer (e.g., controlling the syntax as a style).

This paper tackles the problem of disentangling the latent representations of style and content in language models. We propose a simple yet effective approach, which incorporates auxiliary multi-task and adversarial objectives, for style prediction and bag-of-words prediction, respectively. We show, both qualitatively and quantitatively, that the style and content are indeed disentangled in the latent space. This disentangled latent representation learning can be applied to style transfer on non-parallel corpora. We achieve high performance in terms of transfer accuracy, content preservation, and language fluency, in comparison to various previous approaches.

Variational auto-encoders (VAEs) are widely used in natural language generation due to the regularization of the latent space. However, generating sentences from the continuous latent space does not explicitly model the syntactic information. In this paper, we propose to generate sentences from disentangled syntactic and semantic spaces. Our proposed method explicitly models syntactic information in the VAE’s latent space by using the linearized tree sequence, leading to better performance of language generation. Additionally, the advantage of sampling in the disentangled syntactic and semantic latent spaces enables us to perform novel applications, such as the unsupervised paraphrase generation and syntax transfer generation. Experimental results show that our proposed model achieves similar or better performance in various tasks, compared with state-of-the-art related work.

The variational autoencoder (VAE) imposes a probabilistic distribution (typically Gaussian) on the latent space and penalizes the Kullback-Leibler (KL) divergence between the posterior and prior. In NLP, VAEs are extremely difficult to train due to the problem of KL collapsing to zero. One has to implement various heuristics such as KL weight annealing and word dropout in a carefully engineered manner to successfully train a VAE for text. In this paper, we propose to use the Wasserstein autoencoder (WAE) for probabilistic sentence generation, where the encoder could be either stochastic or deterministic. We show theoretically and empirically that, in the original WAE, the stochastically encoded Gaussian distribution tends to become a Dirac-delta function, and we propose a variant of WAE that encourages the stochasticity of the encoder. Experimental results show that the latent space learned by WAE exhibits properties of continuity and smoothness as in VAEs, while simultaneously achieving much higher BLEU scores for sentence reconstruction.

The variational encoder-decoder (VED) encodes source information as a set of random variables using a neural network, which in turn is decoded into target data using another neural network. In natural language processing, sequence-to-sequence (Seq2Seq) models typically serve as encoder-decoder networks. When combined with a traditional (deterministic) attention mechanism, the variational latent space may be bypassed by the attention model, and thus becomes ineffective. In this paper, we propose a variational attention mechanism for VED, where the attention vector is also modeled as Gaussian distributed random variables. Results on two experiments show that, without loss of quality, our proposed method alleviates the bypassing phenomenon as it increases the diversity of generated sentences.

The paper presents a system for locating a pun word. The developed method calculates a score for each word in a pun, using a number of components, including its Inverse Document Frequency (IDF), Normalized Pointwise Mutual Information (NPMI) with other words in the pun text, its position in the text, part-of-speech and some syntactic features. The method achieved the best performance in the Heterographic category and the second best in the Homographic. Further analysis showed that IDF is the most useful characteristic, whereas the count of words with which the given word has high NPMI has a negative effect on performance.

This paper describes our system for subtask-A: SDQC for RumourEval, task-8 of SemEval 2017. Identifying rumours, especially for breaking news events as they unfold, is a challenging task due to the absence of sufficient information about the exact rumour stories circulating on social media. Determining the stance of Twitter users towards rumourous messages could provide an indirect way of identifying potential rumours. The proposed approach makes use of topic independent features from two categories, namely cue features and message specific features to fit a gradient boosting classifier. With an accuracy of 0.78, our system achieved the second best performance on subtask-A of RumourEval.

This paper discusses the approach taken by the UWaterloo team to arrive at a solution for the Fine-Grained Sentiment Analysis problem posed by Task 5 of SemEval 2017. The paper describes the document vectorization and sentiment score prediction techniques used, as well as the design and implementation decisions taken while building the system for this task. The system uses text vectorization models, such as N-gram, TF-IDF and paragraph embeddings, coupled with regression model variants to predict the sentiment scores. Amongst the methods examined, unigrams and bigrams coupled with simple linear regression obtained the best baseline accuracy. The paper also explores data augmentation methods to supplement the training dataset. This system was designed for Subtask 2 (News Statements and Headlines).

This paper describes the UWaterloo affect prediction system developed for EmoInt-2017. We delve into our feature selection approach for affect intensity, affect presence, sentiment intensity and sentiment presence lexica alongside pre-trained word embeddings, which are utilized to extract emotion intensity signals from tweets in an ensemble learning approach. The system employs emotion specific model training, and utilizes distinct models for each of the emotion corpora in isolation. Our system utilizes gradient boosted regression as the primary learning technique to predict the final emotion intensities.