Table 5 Alignment statistic comparison of aci-validation with a sample of real doctor-patient.

From: Aci-bench: a Novel Ambient Clinical Intelligence Dataset for Benchmarking Automatic Visit Note Generation

 

consult

aci-corpus

dialogue

  

avg length (no speaker tokens) (tok)

1505

1203

avg length (sentences)

141

80

note

  

avg length (tok)

683

492

avg length (sentences)

66

49

annotation

  

fraction note sentences aligned

0.84

0.95

fraction transcript sentences aligned

0.34

0.49

fraction crossing annotations

0.67

0.75

avg alignment text similarity

0.15

0.12

avg encounter dialogue-note text similarity

0.26

0.31

% note sentences with labels

  

DICTATION

8

4

QA

15

43

STATEMENT

23

29

VERBALIZATION/STATEMENT2SCRIBE

17

7

  1. (DICTATION: word-for-word copy-paste statements from the transcript, QA: question-answer conversation adjacency pairs, STATEMENT: conversation statements, VERBALIZATION/STATEMENT2SCRIBE: directed instructions or content to a external scribe).
Back to article page