Predicting the CEFR Level of English Texts

I came across a project here. It’s as the title goes about the same thing EnglishGrammar.Pro is preoccupied with.  I was interested in using his ‘data set’ with our profiler here.  There is always going to be a difference between receptive skills such as reading and productive skills such as writing.  It might also be a case that the texts for A1 students are harder than A1 so that students have something to learn from them.  The same goes for the A2 reading sample which seems to be intentionally B levels. etc.

The first A1 text sample, is predicted to be A2 by our profiler.

I hate my roommate.
I’m sorry to hear that. What’s wrong with her?
A lot of things. First of all, she snores so loudly!
That’s not really her fault. What else?
She never takes a shower.
That’s disgusting! You should tell her to take a shower.
I did! She got mad at me and then proceeded to slap me.
Ouch! Did you try to get a new roommate?
tried, but I could not. No one was willing to switch.
I guess you can get a new room by yourself.
can’t afford it.
Ugh, I hate her so much.
It’s okay. The year will be over before you know it.


The A2 sample is predicted to be B1-B2

Hey, Professor Hill.
What is it?
I’ve been having trouble with derivatives.

That’s not good. There’s a test on derivatives next week.
I know. Can I make an appointment with you for some help?
I’m a busy man. You didn’t even check the schedule yet!
I know I’m busy because many people already made appointments.
So I’m hopeless?
Of course not!
There’s the tutoring center.
But the tutors are not professors.
They’re just students.
They might surprise you.

We won’t post more texts because they are so long.  Each level higher got longer and had much less EVP listed vocabulary. Basically:

  • B1 text was predicted to B1-B2.
  • B2 text was predicted to B2.
  • C1 text was predicted to B2.
  • C2 text was predicted to C1

If the profiler added all the unlisted vocabulary as C2 then this would be a more accurate picture.