This research focuses on the learning of syntactic infor-
mation of natural language. One successful approach to 
learning of syntax searches for substitutable parts in a 
plain text corpus and marks them as possible constituents.
Alignment-Based Learning (ABL) is a system that 
implements this approach and extends it with a probabi-
listic disambiguation phase.
Currently, the main problem with ABL is that finding the 
possible constituents is relatively slow, which 
restricts the system to small corpora. 

The aim of this research is to speed-up the search for 
possible constituents, which will allow handling of large
corpora. We expect that large corpora will improve 
accuracy of the learning system, since not only more pos-
sible constituents can be found, but also more precise 
statistics can be applied in the disambiguation phase. 
We introduce suffix trees and show their merits in this 
task. Preliminary results show that although this 
approach seems to yield lower recall, the learning 
process itself is significantly faster.