演講者：政治大學資訊科學系 黃瀚萱 助理教授
演講題目：Deep Learning and Low Resource Natural Language Understanding
演講摘要：Neural network-based models improve the performances of many
NLP tasks in these years. Learning-based models rely on labeled
data for training and testing. However, one of the major challenges
of advanced NLP tasks is the lack of labeled data, and it is costly
and time-consuming to construct the data by manual annotation.
This talk introduces our recent work on low-resource natural language
understanding, including the tasks of Chinese tense prediction,
irony detection, structure information analysis, and stance detection.
Based on the property of individual task, we present a strategy to
address the issue of data sparsity. The approaches such as cross-lingual
transfer learning and self-labeled data pruning will be discussed.