Spoken propositional idea density, a measure to help second language English speaking students: A multicentre cohort study

Figures & data

Figure 1. Overview of the research methodological flow.

Table 1. Linguistic metrics of lectures used in the study.

Lecture	Words used	Propositions	Idea density	Syllables	Syllables per word	Length (seconds)	Words per minute	Syllables/minute
Materials science A (MID)	281	152	0.541	612	2.18	185	91.1	3.3
Materials science B (LID)	380	160	0.421	731	1.92	240	95	3.1
Polymer chemistry A (LID)	284	164	0.577	558	1.96	187	91.1	2.9
Polymer chemistry B (MID)	356	163	0.458	633	1.78	218	98	2.9

Table 2. Main differences between MPL and RFB.

Characteristics	MLP	RBF
Classification	Separate classes via hyperplanes	Separate classes via hyperspheres
Learning	Distributed learning	Localized learning and train faster
Structure	One or more hidden layers	Only one layer, more hidden neurons

Figure 2. STROBE flow diagram showing study participation numbers.

Table 3. Overall average scores achieved.

Full cohort	MID median (IQR)	LID median (IQR)	p-Value*
75 students	6 (3)	7 (3)	0.049

*Wilcoxon signed rank test.

Figure 3. (A) MLP architecture with output layer neuron, (B) RBF architecture with output layer neurons. Legend: Independent variables: Gender (female, male); Language (first/second language English speaker, Age (<21, 21 and above). Dependent variables: Total score (1 = <70%, 2 = 70% and above).

Figure 4. (A) MPL’s AUROC (0.641), (B) RBF’s AUROC (0.527).

Figure 5. Average MCQ test score difference after MID and LID lectures from L1 and L2 English speakers.

Table A1. MPL and RBF artificial neural network outputs.

Items	MPL	RBF
Sample
Training	53	53
Testing	22	22
Total	75	75
Input layer
1	Language	Language
2	Age	Age
3	Gender	Gender
Hidden layers
Number	2	2
Activation function	Hyperbolic tangent	Softmax
Output layer
Dependent variable	1	1
Number of units	2	2
Activation function	Softmax	Identity
Error function	Cross entropy	Sum of squares
Model summary
Training	Cross entropy error	Sum of squares
Percentage of incorrect predictions	43.4	43.4
Testing	Cross entropy error	Sum of squares
Percentage of incorrect predictions	22.7	40.9
Classification
Training: percentage of correct prediction	56.6	56.6
Testing: percentage of correct prediction	77.3	59.1
AUROC
Below 70%	0.641	0.527
70% and above	0.641	0.527
Importance of independent variables
Language	0.406	0.596
Age	0.378	0.218
Gender	0.216	0.186