Prior work on language model pre-training has explored different architectures and learning objectives, but differences in data, hyperparameters and evaluation make a principled comparison difficult. In this work, we focus on bidirectionality as a key factor that differentiates existing approaches, and present a comprehensive study of its role in next token prediction, text infilling, zero-shot priming and fine-tuning. We propose a new framework that generalizes prior approaches, including fully unidirectional models like GPT, fully bidirectional models like BERT, and hybrid models like CM3 and prefix LM. Our framework distinguishes between two notions of bidirectionality (bidirectional context and bidirectional attention) and allows us to control each of them separately. We find that the optimal configuration is largely application-dependent (e.g., bidirectional attention is beneficial for fine-tuning and infilling, but harmful for next token prediction and zero-shot priming). We train models with up to 6.7B parameters, and find differences to remain consistent at scale. While prior work on scaling has focused on left-to-right autoregressive models, our results suggest that this approach comes with some trade-offs, and it might be worthwhile to develop very large bidirectional models.
Contrastive vision-language models like CLIP have shown great progress in zero-shot transfer learning. This new paradigm uses large-scale image-text pairs for training and aligns images and texts in a common embedding space. In the inference stage, the proper text description, known as prompt, needs to be carefully designed for zero-shot transfer. To avoid laborious prompt engineering and simultaneously improve transfer performance, recent works such as CoOp, CLIP-Adapter and Tip-Adapter propose to adapt vision-language models for downstream image recognition tasks by either optimizing the continuous prompt representations or training an additional adapter network on top of the pre-trained vision-language models on a small set of labeled data. Though promising improvements are achieved, using labeled images from target datasets may violate the intention of zero-shot transfer of pre-trained vision-language models. In this paper, we propose an unsupervised prompt learning (UPL) framework, which does not require any annotations of the target dataset, to improve the zero-shot transfer of CLIP-like vision-language models. Experimentally, for zero-shot transfer, our UPL outperforms original CLIP with prompt engineering and on ImageNet as well as other 10 datasets. An enhanced version of UPL is even on par with the 8-shot CoOp and the 8-shot TIP-Adapter on most datasets while our method does not need any labeled images for training. Code and models are available at https://github.com/tonyhuang2022/UPL.
Machine learning has been utilized to perform tasks in many different domains such as classification, object detection, image segmentation and natural language analysis. Data labeling has always been one of the most important tasks in machine learning. However, labeling large amounts of data increases the monetary cost in machine learning. As a result, researchers started to focus on reducing data annotation and labeling costs. Transfer learning was designed and widely used as an efficient approach that can reasonably reduce the negative impact of limited data, which in turn, reduces the data preparation cost. Even transferring previous knowledge from a source domain reduces the amount of data needed in a target domain. However, large amounts of annotated data are still demanded to build robust models and improve the prediction accuracy of the model. Therefore, researchers started to pay more attention on auto annotation and labeling. In this survey paper, we provide a review of previous techniques that focuses on optimized data annotation and labeling for video, audio, and text data.
Stance detection infers a text author's attitude towards a target. This is challenging when the model lacks background knowledge about the target. Here, we show how background knowledge from Wikipedia can help enhance the performance on stance detection. We introduce Wikipedia Stance Detection BERT (WS-BERT) that infuses the knowledge into stance encoding. Extensive results on three benchmark datasets covering social media discussions and online debates indicate that our model significantly outperforms the state-of-the-art methods on target-specific stance detection, cross-target stance detection, and zero/few-shot stance detection.
In recent years, the problem of misinformation on the web has become widespread across languages, countries, and various social media platforms. Although there has been much work on automated fake news detection, the role of images and their variety are not well explored. In this paper, we investigate the roles of image and text at an earlier stage of the fake news detection pipeline, called claim detection. For this purpose, we introduce a novel dataset, MM-Claims, which consists of tweets and corresponding images over three topics: COVID-19, Climate Change and broadly Technology. The dataset contains roughly 86000 tweets, out of which 3400 are labeled manually by multiple annotators for the training and evaluation of multimodal models. We describe the dataset in detail, evaluate strong unimodal and multimodal baselines, and analyze the potential and drawbacks of current models.
Accuracy of English-language Question Answering (QA) systems has improved significantly in recent years with the advent of Transformer-based models (e.g., BERT). These models are pre-trained in a self-supervised fashion with a large English text corpus and further fine-tuned with a massive English QA dataset (e.g., SQuAD). However, QA datasets on such a scale are not available for most of the other languages. Multi-lingual BERT-based models (mBERT) are often used to transfer knowledge from high-resource languages to low-resource languages. Since these models are pre-trained with huge text corpora containing multiple languages, they typically learn language-agnostic embeddings for tokens from different languages. However, directly training an mBERT-based QA system for low-resource languages is challenging due to the paucity of training data. In this work, we augment the QA samples of the target language using translation and transliteration into other languages and use the augmented data to fine-tune an mBERT-based QA model, which is already pre-trained in English. Experiments on the Google ChAII dataset show that fine-tuning the mBERT model with translations from the same language family boosts the question-answering performance, whereas the performance degrades in the case of cross-language families. We further show that introducing a contrastive loss between the translated question-context feature pairs during the fine-tuning process, prevents such degradation with cross-lingual family translations and leads to marginal improvement. The code for this work is available at https://github.com/gokulkarthik/mucot.
Text style transfer is a hot issue in recent natural language processing,which mainly studies the text to adapt to different specific situations, audiences and purposes by making some changes. The style of the text usually includes many aspects such as morphology, grammar, emotion, complexity, fluency, tense, tone and so on. In the traditional text style transfer model, the text style is generally relied on by experts knowledge and hand-designed rules, but with the application of deep learning in the field of natural language processing, the text style transfer method based on deep learning Started to be heavily researched. In recent years, text style transfer is becoming a hot issue in natural language processing research. This article summarizes the research on the text style transfer model based on deep learning in recent years, and summarizes, analyzes and compares the main research directions and progress. In addition, the article also introduces public data sets and evaluation indicators commonly used for text style transfer. Finally, the existing characteristics of the text style transfer model are summarized, and the future development trend of the text style transfer model based on deep learning is analyzed and forecasted.
The proliferation of news media available online simultaneously presents a valuable resource and significant challenge to analysts aiming to profile and understand social and cultural trends in a geographic location of interest. While an abundance of news reports documenting significant events, trends, and responses provides a more democratized picture of the social characteristics of a location, making sense of an entire corpus to extract significant trends is a steep challenge for any one analyst or team. Here, we present an approach using natural language processing techniques that seeks to quantify how a set of pre-defined topics of interest change over time across a large corpus of text. We found that, given a predefined topic, we can identify and rank sets of terms, or n-grams, that map to those topics and have usage patterns that deviate from a normal baseline. Emergence, disappearance, or significant variations in n-gram usage present a ground-up picture of a topic's dynamic salience within a corpus of interest.
A lot of natural language processing problems need to encode the text sequence as a fix-length vector, which usually involves aggregation process of combining the representations of all the words, such as pooling or self-attention. However, these widely used aggregation approaches did not take higher-order relationship among the words into consideration. Hence we propose a new way of obtaining aggregation weights, called eigen-centrality self-attention. More specifically, we build a fully-connected graph for all the words in a sentence, then compute the eigen-centrality as the attention score of each word. The explicit modeling of relationships as a graph is able to capture some higher-order dependency among words, which helps us achieve better results in 5 text classification tasks and one SNLI task than baseline models such as pooling, self-attention and dynamic routing. Besides, in order to compute the dominant eigenvector of the graph, we adopt power method algorithm to get the eigen-centrality measure. Moreover, we also derive an iterative approach to get the gradient for the power method process to reduce both memory consumption and computation requirement.}
We propose a structured extension to bidirectional-context conditional language generation, or "infilling," inspired by Frame Semantic theory (Fillmore, 1976). Guidance is provided through two approaches: (1) model fine-tuning, conditioning directly on observed symbolic frames, and (2) a novel extension to disjunctive lexically constrained decoding that leverages frame semantic lexical units. Automatic and human evaluations confirm that frame-guided generation allows for explicit manipulation of intended infill semantics, with minimal loss of indistinguishability from the human-generated text. Our methods flexibly apply to a variety of use scenarios, and we provide an interactive web demo available at https://nlp.jhu.edu/demos.