Figure 4

Text input and structured EHR input segmented by PLM tokenizer. In the processing of structured EHR input, “74.3” is split into “74”, decimal point and “3”, “54.5455” is split into “54”, decimal point, “54” and “## 55”, “239.7” is split into “239”, decimal point and “7”. The mapping between the natural language texts and the structured EHRs is difficult to understand.