BioCopy: A Plug-And-Play Span Copy Mechanism in Seq2Seq Models
This addresses a specific limitation in seq2seq models for natural language generation tasks, but it is incremental as it builds on existing copy mechanisms.
The paper tackles the problem of existing copy mechanisms in seq2seq models losing essential tokens when copying long spans by proposing BioCopy, a plug-and-play architecture that uses BIO tags and mask strategies, resulting in outperforming baseline models on two generative tasks.
Copy mechanisms explicitly obtain unchanged tokens from the source (input) sequence to generate the target (output) sequence under the neural seq2seq framework. However, most of the existing copy mechanisms only consider single word copying from the source sentences, which results in losing essential tokens while copying long spans. In this work, we propose a plug-and-play architecture, namely BioCopy, to alleviate the problem aforementioned. Specifically, in the training stage, we construct a BIO tag for each token and train the original model with BIO tags jointly. In the inference stage, the model will firstly predict the BIO tag at each time step, then conduct different mask strategies based on the predicted BIO label to diminish the scope of the probability distributions over the vocabulary list. Experimental results on two separate generative tasks show that they all outperform the baseline models by adding our BioCopy to the original model structure.