Semi-Supervised Training Using Adversarial Multi-Task Learning for Spoken Language Understanding

  • Advised by Prof. Kai Yu;

  • Published on IEEE International Conference on Acoustics, Speech, and Signal Processing, 2018. Paper;

  • Designed a multitask learning method by neatly merging a unidirectional / bidirectional language model and a slot tagging model in SLU;

  • Exploited a shared space for both tasks and two task-specific spaces for classification, extracted generalized language knowledge, and regularized the slot tagging model;

  • Developed an adversarial discriminator to obtain more task-independent sharing information for SLU;

  • Implemented in PyTorch;

  • Achieved the best performance in slot tagging task on both small (ATIS) and large-scale datasets (e.g. improved F1-score from 67.66% to 71.55% with 5k labeled sentences).

Leave a Comment

Your email address will not be published. Required fields are marked *