2020/05/22 by Aakanksha Naik, Naik, Aakanksha, Carolyn Penstein Rosé +1
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2005.11355
openalex publication_date 2020/05/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We tackle the task of building supervised event trigger identification models\nwhich can generalize better across domains. Our work leverages the adversarial\ndomain adaptation (ADA) framework to introduce domain-invariance. ADA uses\nadversarial training to construct representations that are predictive for\ntrigger identification, but not predictive of the example's domain. It requires\nno labeled data from the target domain, making it completely unsupervised.\nExperiments with two domains (English literature and news) show that ADA leads\nto an average F1 score improvement of 3.9 on out-of-domain data. Our best\nperforming model (BERT-A) reaches 44-49 F1 across both domains, using no\nlabeled target data. Preliminary experiments reveal that finetuning on 1%\nlabeled data, followed by self-training leads to substantial improvement,\nreaching 51.5 and 67.2 F1 on literature and news respectively.\n