Ngram counts are sometimes wrong

[firemuzzy]

I have noticed that for some phrases the ngram counts are greater than the number of occurrences in the text.

Here is a an example https://runkit.com/firemuzzy/5af1d67875914d001263570c

If you look at the phrase free for, the provided text has 2 occurrences, but nlp-compromise returns a count of 3. Coincidentally the phrase free occurs 3 times, I wonder if something is messing up in the normalization.

At the same time the phrase just not user friendly only occurs 1 time, but nlp-compromise reports it with a count of 2. That discrepancy completely puzzles me.

Am I missing something with how ngrams works?

[spencermountain]

hey Michael, nice find.
I'll take a look at fixing this this week.
thanks

[spencermountain]

ha! oh wow, it's the contraction - It's free for.
it's creating the gram [is] free for.
this is a great bug, will fix it today.
cheers

[spencermountain]

fixed in 11.8.0
thanks!