As the excessive pre-training cost arouses the need to improve efficiency, considerable efforts have been made to train BERT progressively–start from an inferior but low-cost model and gradually increase the computational complexity. Our objective is to help advance the understanding of such Transformer growth and discover principles that guide progressive training. First, we find that similar to network architecture selection, Transformer growth also favors compound scaling. Specifically, while existing methods only conduct network growth in a single dimension, we observe that it is beneficial to use compound growth operators and balance multiple dimensions (e.g., depth, width, and input length of the model). Moreover, we explore alternative growth operators in each dimension via controlled comparison to give practical guidance for operator selection. In light of our analyses, the proposed method CompoundGrow speeds up BERT pre-training by 73.6% and 82.2% for the base and large models respectively while achieving comparable performances.

Large pre-trained transformer-based language models have achieved impressive results on a wide range of NLP tasks. In the past few years, Knowledge Distillation(KD) has become a popular paradigm to compress a computationally expensive model to a resource-efficient lightweight model. However, most KD algorithms, especially in NLP, rely on the accessibility of the original training dataset, which may be unavailable due to privacy issues. To tackle this problem, we propose a novel two-stage data-free distillation method, named Adversarial self-Supervised Data-Free Distillation (AS-DFD), which is designed for compressing large-scale transformer-based models (e.g., BERT). To avoid text generation in discrete space, we introduce a Plug & Play Embedding Guessing method to craft pseudo embeddings from the teacher’s hidden knowledge. Meanwhile, with a self-supervised module to quantify the student’s ability, we adapt the difficulty of pseudo embeddings in an adversarial training manner. To the best of our knowledge, our framework is the first data-free distillation framework designed for NLP tasks. We verify the effectiveness of our method on several text classification datasets.

In this paper, we study a new task of synonym expansion using transitivity, and propose a novel approach named SynET, which considers both the contexts of two given synonym pairs. It introduces an auxiliary task to reduce the impact of noisy sentences, and proposes a Multi-Perspective Entity Matching Network to match entities from multiple perspectives. Extensive experiments on a real-world dataset show the effectiveness of our approach.

Paraphrases are important linguistic resources for a wide variety of NLP applications. Many techniques for automatic paraphrase mining from general corpora have been proposed. While these techniques are successful at discovering generic paraphrases, they often fail to identify domain-specific paraphrases (e.g., staff, concierge in the hospitality domain). This is because current techniques are often based on statistical methods, while domain-specific corpora are too small to fit statistical methods. In this paper, we present an unsupervised graph-based technique to mine paraphrases from a small set of sentences that roughly share the same topic or intent. Our system, Essentia, relies on word-alignment techniques to create a word-alignment graph that merges and organizes tokens from input sentences. The resulting graph is then used to generate candidate paraphrases. We demonstrate that our system obtains high quality paraphrases, as evaluated by crowd workers. We further show that the majority of the identified paraphrases are domain-specific and thus complement existing paraphrase databases.

Compared to entity coreference resolution, there is a relatively small amount of work on event coreference resolution. Much work on event coreference was done for English. In fact, to our knowledge, there are no publicly available results on Chinese event coreference resolution. This paper describes the design, implementation, and evaluation of SinoCoreferencer, an end-to-end state-of-the-art ACE-style Chinese event coreference system. We have made SinoCoreferencer publicly available, in hope to facilitate the development of high-level Chinese natural language applications that can potentially benefit from event coreference information.