Improving Vietnamese Named Entity Recognition from Speech Using Word Capitalization and Punctuation Recovery Models

Studies on the Named Entity Recognition (NER) task have shown outstanding\nresults that reach human parity on input texts with correct text formattings,\nsuch as with proper punctuation and capitalization. However, such conditions\nare not available in applications where the input is speech, because the text\nis generated from a speech recognition system (ASR), and that the system does\nnot consider the text formatting. In this paper, we (1) presented the first\nVietnamese speech dataset for NER task, and (2) the first pre-trained public\nlarge-scale monolingual language model for Vietnamese that achieved the new\nstate-of-the-art for the Vietnamese NER task by 1.3% absolute F1 score\ncomparing to the latest study. And finally, (3) we proposed a new pipeline for\nNER task from speech that overcomes the text formatting problem by introducing\na text capitalization and punctuation recovery model (CaPu) into the pipeline.\nThe model takes input text from an ASR system and performs two tasks at the\nsame time, producing proper text formatting that helps to improve NER\nperformance. Experimental results indicated that the CaPu model helps to\nimprove by nearly 4% of F1-score.\n