Masked language models (MLMs) conventionally mask 15% of tokens due to the belief that more masking would leave insufficient context to learn good representations; this masking rate has been widely used, regardless of model sizes or masking strategies. In this work, we revisit this important choice of MLM pre-training. We first establish that 15% is not universally optimal, and larger models should adopt a higher masking rate. Specifically, we find that masking 40% outperforms 15% for BERT-large size models on GLUE and SQuAD. Interestingly, an extremely high masking rate of 80% can still preserve 95% fine-tuning performance and most of the accuracy in linguistic probing, challenging the conventional wisdom about the role of the masking rate. We then examine the interplay between masking rates and masking strategies and find that uniform masking requires a higher masking rate compared to sophisticated masking strategies such as span or PMI masking. Finally, we argue that increasing the masking rate has two distinct effects: it leads to more corruption, which makes the prediction task more difficult; it also enables more predictions, which benefits optimization. Using this framework, we revisit BERT's 80-10-10 corruption strategy. Together, our results contribute to a better understanding of MLM pre-training.
This paper presents Masked ELMo, a new RNN-based model for language model pre-training, evolved from the ELMo language model. Contrary to ELMo which only uses independent left-to-right and right-to-left contexts, Masked ELMo learns fully bidirectional word representations. To achieve this, we use the same Masked language model objective as BERT. Additionally, thanks to optimizations on the LSTM neuron, the integration of mask accumulation and bidirectional truncated backpropagation through time, we have increased the training speed of the model substantially. All these improvements make it pos…
In this paper, we study how to use masked signal modeling in vision and language (V+L) representation learning. Instead of developing masked language modeling (MLM) and masked image modeling (MIM) independently, we propose to build joint masked vision and language modeling, where the masked signal of one modality is reconstructed with the help from another modality. This is motivated by the nature of image-text paired data that both of the image and the text convey almost the same information but in different formats. The masked signal reconstruction of one modality conditioned on another moda…
The Hebrew University of Jerusalem（耶路撒冷希伯来大学）；Ben Gurion University（本·古里安大学）
Masked language modeling (MLM) is one of the key sub-tasks in vision-language pretraining. In the cross-modal setting, tokens in the sentence are masked at random, and the model predicts the masked tokens given the image and the text. In this paper, we observe several key disadvantages of MLM in this setting. First, as captions tend to be short, in a third of the sentences no token is sampled. Second, the majority of masked tokens are stop-words and punctuation, leading to under-utilization of the image. We investigate a range of alternative masking strategies specific to the cross-modal setti…
Masked language model and autoregressive language model are two types of language models. While pretrained masked language models such as BERT overwhelm the line of natural language understanding (NLU) tasks, autoregressive language models such as GPT are especially capable in natural language generation (NLG). In this paper, we propose a probabilistic masking scheme for the masked language model, which we call probabilistically masked language model (PMLM). We implement a specific PMLM with a uniform prior distribution on the masking ratio named u-PMLM. We prove that u-PMLM is equivalent to a…
Mask and You Shall Receive: Optimizing Masked Language Modeling For Pretraining BabyLMs
School of Computation, Information and Technology, TU Munich（计算、信息与技术学院，慕尼黑工业大学）；Munich Center for Machine Learning（慕尼黑机器学习中心）；Munich Data Science Institute（慕尼黑数据科学研究所）
Korea University（高丽大学）；The Catholic University of Korea（韩国天主教大学）
Masked language modeling (MLM) has been widely used for pre-training effective bidirectional representations, but incurs substantial training costs. In this paper, we propose a novel concept-based curriculum masking (CCM) method to efficiently pre-train a language model. CCM has two key differences from existing curriculum learning approaches to effectively reflect the nature of MLM. First, we introduce a carefully-designed linguistic difficulty criterion that evaluates the MLM difficulty of each token. Second, we construct a curriculum that gradually masks words related to the previously mask…
Most machine translation systems generate text autoregressively from left to right. We, instead, use a masked language modeling objective to train a model to predict any subset of the target words, conditioned on both the input text and a partially masked target translation. This approach allows for efficient iterative decoding, where we first predict all of the target words non-autoregressively, and then repeatedly mask out and regenerate the subset of words that the model is least confident about. By applying this strategy for a constant number of iterations, our model improves state-of-the-…
Exploring Gradient-Guided Masked Language Model to Detect Textual Adversarial Attacks
School of Information and Communication Technology, Griffith University（格里菲斯大学信息与通信技术学院）；School of Computer Science, University of Technology Sydney（悉尼科技大学计算机科学学院）；College of Computer and Information Science, Southwest University（西南大学计算机与信息科学学院）；School of Cyber Science and Engineering, Huazhong University of Science and Technology（华中科技大学网络空间科学与工程学院）
Decompilation is the procedure of transforming binary programs into a high-level representation, such as source code, for human analysts to examine. While modern decompilers can reconstruct and recover much information that is discarded during compilation, inferring variable names is still extremely difficult. Inspired by recent advances in natural language processing, we propose a novel solution to infer variable names in decompiled code based on Masked Language Modeling, Byte-Pair Encoding, and neural architectures such as Transformers and BERT. Our solution takes \textit{raw} decompiler out…
Kempelen Institute of Intelligent Technologies（凯普伦智能技术研究院）；Gerulata Technologies（Gerulata技术公司）
We introduce a new Slovak masked language model called SlovakBERT. This is to our best knowledge the first paper discussing Slovak transformers-based language models. We evaluate our model on several NLP tasks and achieve state-of-the-art results. This evaluation is likewise the first attempt to establish a benchmark for Slovak language models. We publish the masked language model, as well as the fine-tuned models for part-of-speech tagging, sentiment analysis and semantic textual similarity.
Masked language models have recently been interpreted as energy-based sequence models that can be generated from using a Metropolis--Hastings sampler. This short paper demonstrates how this can be instrumentalized for constrained composition and explores the poetics implied by such a usage. Our focus on constraints makes it especially apt to understand the generated text through the poetics of the OuLiPo movement.
Microsoft Research（微软研究院）；Harbin Institute of Technology（哈尔滨工业大学）
We propose to pre-train a unified language model for both autoencoding and partially autoregressive language modeling tasks using a novel training procedure, referred to as a pseudo-masked language model (PMLM). Given an input text with masked tokens, we rely on conventional masks to learn inter-relations between corrupted tokens and context via autoencoding, and pseudo masks to learn intra-relations between masked spans via partially autoregressive modeling. With well-designed position embeddings and self-attention masks, the context encodings are reused to avoid redundant computation. Moreov…
Beijing University of Posts and Telecommunications（北京邮电大学）；Huawei Technologies Ltd. Co.（华为技术有限公司）
Despite pre-training's progress in many important NLP tasks, it remains to explore effective pre-training strategies for dense retrieval. In this paper, we propose RetroMAE, a new retrieval oriented pre-training paradigm based on Masked Auto-Encoder (MAE). RetroMAE is highlighted by three critical designs. 1) A novel MAE workflow, where the input sentence is polluted for encoder and decoder with different masks. The sentence embedding is generated from the encoder's masked input; then, the original sentence is recovered based on the sentence embedding and the decoder's masked input via masked …
Indian Institute of Technology Hyderabad（印度理工学院海得拉巴分校）
Text Style Transfer (TST) is performable through approaches such as latent space disentanglement, cycle-consistency losses, prototype editing etc. The prototype editing approach, which is known to be quite successful in TST, involves two key phases a) Masking of source style-associated tokens and b) Reconstruction of this source-style masked sentence conditioned with the target style. We follow a similar transduction method, in which we transpose the more difficult direct source to target TST task to a simpler Style-Masked Language Model (SMLM) Task, wherein, similar to BERT \cite{bert}, the g…
Tokyo Institute of Technology（东京工业大学）；University of Liverpool（利物浦大学）；Amazon（亚马逊）
Masked Language Models (MLMs) have shown superior performances in numerous downstream NLP tasks when used as text encoders. Unfortunately, MLMs also demonstrate significantly worrying levels of social biases. We show that the previously proposed evaluation metrics for quantifying the social biases in MLMs are problematic due to following reasons: (1) prediction accuracy of the masked tokens itself tend to be low in some MLMs, which raises questions regarding the reliability of the evaluation metrics that use the (pseudo) likelihood of the predicted tokens, and (2) the correlation between the p…
Masked Diffusion Language Models with Frequency-Informed Training
National Technical University of Athens（国家技术大学雅典）；Archimedes RU, Athena RC（阿基米德RU，雅典RC）；University of Bern（伯恩大学）；Institute of Language and Signal Processing, Athena RC（语言与信号处理研究所，雅典RC）
Microsoft Dynamics 365 AI（微软Dynamics 365人工智能）；Microsoft Research（微软研究院）
Masked Language Model (MLM) framework has been widely adopted for self-supervised language pre-training. In this paper, we argue that randomly sampled masks in MLM would lead to undesirably large gradient variance. Thus, we theoretically quantify the gradient variance via correlating the gradient covariance with the Hamming distance between two different masks (given a certain text sequence). To reduce the variance due to the sampling of masks, we propose a fully-explored masking strategy, where a text sequence is divided into a certain number of non-overlapping segments. Thereafter, the token…
University of Amsterdam（阿姆斯特丹大学）；RWTH Aachen University（亚琛工业大学）；eBay（易贝）
The pre-training of masked language models (MLMs) consumes massive computation to achieve good results on downstream NLP tasks, resulting in a large carbon footprint. In the vanilla MLM, the virtual tokens, [MASK]s, act as placeholders and gather the contextualized information from unmasked tokens to restore the corrupted information. It raises the question of whether we can append [MASK]s at a later layer, to reduce the sequence length for earlier layers and make the pre-training more efficient. We show: (1) [MASK]s can indeed be appended at a later layer, being disentangled from the word emb…
