They took IQ tests before and after and gained 12 IQ points after the training. A control group also took the tests before and after but did not receive training, and did not improve. The sample sizes are small, but the effect sizes might be large enough to justify it. They give a p value of 0.008.

The second sentence surprises me a little--there should be training effects increasing the tested IQ of the control group if only 9 weeks passed. That's some evidence for this being luck--if your control group gets unlucky and your experimental group gets lucky, then you see a huge effect.

I want to know if it would be better to try letter level associations or word level ones.

There are 26 letters, but... lots of words.