CLIP evaluation

1 corpus clip scores

  • data: clip_results.parquet (Ivan Nenchev)
  • additional:
    • lexical diversity score
    • 1stPPronoun rate
    • exploded clip scores, not mean!

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: cl_score ~ group + (1 | TN) + (1 | text_chunk) + ld + fstPPr_rate
   Data: dff4

REML criterion at convergence: -406293.2

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-4.7324 -0.5891 -0.0624  0.4691  6.2275 

Random effects:
 Groups     Name        Variance  Std.Dev.
 text_chunk (Intercept) 1.631e-03 0.040381
 TN         (Intercept) 7.633e-05 0.008737
 Residual               1.476e-03 0.038419
Number of obs: 113900, groups:  text_chunk, 3653; TN, 29

Fixed effects:
             Estimate Std. Error        df t value Pr(>|t|)    
(Intercept)  0.234962   0.010406 27.942796  22.579  < 2e-16 ***
groupscz     0.023571   0.004146 28.049269   5.685 4.26e-06 ***
ld          -0.026691   0.034733 36.448483  -0.768    0.447    
fstPPr_rate -0.189069   0.122592 24.946504  -1.542    0.136    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) grpscz ld    
groupscz    -0.217              
ld          -0.731  0.280       
fstPPr_rate -0.696 -0.261  0.105