Improving Natural Language Parser Accuracy by Unknown Word Replacement
Raihan Kibria, Khandaker Tabin Hasan · 2020
Natural language parsers are the basis for further understanding the content written in natural language. Parsers for natural language have been shown to be effective in many NLP tasks, such as, machine translation, sentiment analysis and classification of documents. The existing state-of-the-art parsers, such as Charniak [9], Collins [11], Stanford, OpenNLP, have been shown to have F Score ranging from 85 to 92 percent. The accuracy of the parsers is hampered to a major extent by unknown and unseen words. In this paper we show a novel method on improving the accuracy by incorporating knowledge about the unknown words from external source. Experimental results show our technique improves accuracy. The improvement depends on number of known words present in the model during training. We show that we achieve above one percent improvement on some parsers.