VarMAE: Pre-training of Variational Masked Autoencoder for Domain-adaptive Language Understanding

AbstractPre-trained language models have been widely applied to standard benchmarks. Due to the flexibility of natural language, the available resources in a certain domain can be restricted to support obtaining precise representation. To address this issue, we propose a novel Transformer-based language model named VarMAE for domain-adaptive language understanding. Under the masked autoencoding objective, we design a context uncertainty learning module to encode the token’s context into a smooth latent distribution. The module can produce diverse and well-formed contextual representations. Experiments on science- and finance-domain NLU tasks demonstrate that VarMAE can be efficiently adapted to new domains with limited resources.