Supplementary figure 1. Ablanullon study, each row represents the ablated layer and each column the module that is ablated from that layer, for example the first panel shows ablanullon of anullennullon - key in layer 5. Different layers in GPT2 - XL model were ablated and the consequence of ablanullon on curvature measured for 2000 sentences in UD corpus. Red bar shows the layer where ablanullon was applied. AB Supplementary figure 3. A. curvature values for sampled 2000 sentence in RWKV model ( RNN) for both trained an untrained version. B correlanullon between model generated surprisal and curvature in RWKV model. Diamonds: syntacnullc surprisal Supplementary figure 5: E ffect of different decoding strategies in GPT2 - XL sequence generanullon and its comparison to ground - truth(true) same as figure 4b in the main manuscript.

Influence functions (IF) have been seen as a technique for explaining model predictions through the lens of the training data.

This trend has motivated research on efficient training algorithms designed to improve training, validation, and downstream performance faster than standard training.

Large pre-trained transformers are $\textit{show-stealer}$ in modern-day deep learning, and it becomes crucial to comprehend the parsimonious patterns that exist within them as they grow in scale. With exploding parameter counts, Lottery Ticket Hypothesis (LTH) and its variants, have lost their pragmatism in sparsifying them due to high computation and memory bottleneck of repetitive $\textit{train-prune-retrain}$ routine of iterative magnitude pruning (IMP) which worsens with increasing model size. In this paper, we comprehensively study $\textit{induced sparse patterns}$ across multiple large pre-trained vision and language transformers. We propose the existence of -- $\textbf{essential sparsity}$ defined with a $\textbf{sharp dropping point}$ beyond which the performance declines much faster w.r.t the rise of sparsity level, when we directly remove weights with the smallest magnitudes in $\textbf{one-shot}$. We also present an intriguing emerging phenomenon of $\textbf{abrupt sparsification}$ during the pre-training of BERT, i.e., BERT suddenly becomes heavily sparse in pre-training after certain iterations. Moreover, our observations also indicate a $\textbf{counter-intuitive}$ finding that BERT trained with a larger amount of pre-training data tends to have a better ability to condense knowledge in comparatively relatively fewer parameters. Lastly, we investigate the effect of the pre-training loss on essential sparsity and discover that self-supervised learning (SSL) objectives trigger stronger emergent sparsification properties than supervised learning (SL). All our codes will be publicly available.

With the capability of modeling bidirectional contexts, denoising autoencoding based pretraining like BERT achieves better performance than pretraining approaches based on autoregressive language modeling. However, relying on corrupting the input with masks, BERT neglects dependency between the masked positions and suffers from a pretrain-finetune discrepancy. In light of these pros and cons, we propose XLNet, a generalized autoregressive pretraining method that (1) enables learning bidirectional contexts by maximizing the expected likelihood over all permutations of the factorization order and (2) overcomes the limitations of BERT thanks to its autoregressive formulation.

Neural networks models for NLP are typically implemented without the explicit encoding of language rules and yet they are able to break one performance record after another. This has generated a lot of research interest in interpreting the representations learned by these networks. We propose here a novel interpretation approach that relies on the only processing system we have that does understand language: the human brain. We use brain imaging recordings of subjects reading complex natural text to interpret word and sequence embeddings from 4 recent NLP models - ELMo, USE, BERT and Transformer-XL. We study how their representations differ across layer depth, context length, and attention type.

It is intuitive that NLP tasks for logographic languages like Chinese should benefit from the use of the glyph information in those languages. However, due to the lack of rich pictographic evidence in glyphs and the weak generalization ability of standard computer vision models on character data, an effective way to utilize the glyph information remains to be found. In this paper, we address this gap by presenting Glyce, the glyph-vectors for Chinese character representations. We make three major innovations: (1) We use historical Chinese scripts (e.g., bronzeware script, seal script, traditional Chinese, etc) to enrich the pictographic evidence in characters; (2) We design CNN structures (called tianzege-CNN) tailored to Chinese character image processing; and (3) We use image-classification as an auxiliary task in a multi-task learning setup to increase the model's ability to generalize. We show that glyph-based models are able to consistently outperform word/char ID-based models in a wide range of Chinese NLP tasks. When combing with BERT, we are able to set new state-of-the-art results for a variety of Chinese NLP tasks, including language modeling, tagging (NER, CWS, POS), sentence pair classification (BQ, LCQMC, XNLI, NLPCC-DBQA), single sentence classification tasks (ChnSentiCorp, the Fudan corpus, iFeng), dependency parsing, and semantic role labeling. For example, the proposed model achieves an F1 score of 81.6 on the OntoNotes dataset of NER, +1.5 over BERT; it achieves an almost perfect accuracy of 99.8\% on the the Fudan corpus for text classification.

Progress in natural language processing (NLP) models that estimate representations of word sequences has recently been leveraged to improve the understanding of language processing in the brain. However, these models have not been specifically designed to capture the way the brain represents language meaning. We hypothesize that fine-tuning these models to predict recordings of brain activity of people reading text will lead to representations that encode more brain-activity-relevant language information. We demonstrate that a version of BERT, a recently introduced and powerful language model, can improve the prediction of brain activity after fine-tuning. We show that the relationship between language and brain activity learned by BERT during this fine-tuning transfers across multiple participants. We also show that, for some participants, the fine-tuned representations learned from both magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) are better for predicting fMRI than the representations learned from fMRI alone, indicating that the learned representations capture brain-activity-relevant information that is not simply an artifact of the modality. While changes to language representations help the model predict brain activity, they also do not harm the model's ability to perform downstream NLP tasks. Our findings are notable for research on language understanding in the brain.

BERT adopts masked language modeling (MLM) for pre-training and is one of the most successful pre-training models. Since BERT neglects dependency among predicted tokens, XLNet introduces permuted language modeling (PLM) for pre-training to address this problem. However, XLNet does not leverage the full position information of a sentence and thus suffers from position discrepancy between pre-training and fine-tuning. In this paper, we propose MPNet, a novel pre-training method that inherits the advantages of BERT and XLNet and avoids their limitations. MPNet leverages the dependency among predicted tokens through permuted language modeling (vs. MLM in BERT), and takes auxiliary position information as input to make the model see a full sentence and thus reducing the position discrepancy (vs.

Pre-trained language models like BERT and its variants have recently achieved impressive performance in various natural language understanding tasks. However, BERT heavily relies on the global self-attention block and thus suffers large memory footprint and computation cost. Although all its attention heads query on the whole input sequence for generating the attention map from a global perspective, we observe some heads only need to learn local dependencies, which means existence of computation redundancy. We therefore propose a novel span-based dynamic convolution to replace these self-attention heads to directly model local dependencies. The novel convolution heads, together with the rest self-attention heads, form a new mixed attention block that is more efficient at both global and local context learning. We equip BERT with this mixed attention design and build a ConvBERT model. Experiments have shown that ConvBERT significantly outperforms BERT and its variants in various downstream tasks, with lower training cost and fewer model parameters. Remarkably, ConvBERTbase model achieves 86.4 GLUE score, 0.7 higher than ELECTRAbase, using less than 1/4 training cost. Code and pre-trained models will be released.