Measuring how a child reads aloud, and showing the teacher the working.
READSmart listens to a pupil read a passage aloud, measures reading speed, and corrects for the ways a machine mishears a child. Teachers keep the judgement.
Read the findingsWhat calibration changed
The system used to under-read children. Calibration removed that systematic offset, in grouped cross-validation of 167 recordings from 42 pupils.
−13.603
mean bias, before calibration (WCPM)
0.149
after calibration (WCPM)
It did not make individual scores agree with a teacher. Average error remains 10.043 WCPM, and a single score falls within −30.068 to +30.366 WCPM of a teacher’s — about 3.2× wider than two teachers differ from each other (±9.553). Correcting the average is not the same as being right about one child.
A screening aid. The teacher remains the authority.
How we measured this