Alignment-Based Learning (ABL) is a symbolic grammar inference
framework that has been applied successfully to several unsupervised
machine learning tasks in Natural Language Processing (NLP).  Given
sequences of symbols, ABL induces structure by aligning and comparing
the input sequences.  Regularities in the input sequences are used to
assign structure to the sequences.

ABL consists of three phases.  The first phase, alignment
learning, builds a search space of possible structures, called
hypotheses, by comparing the sequences.  The second phase,
called clustering clusters subsequences that share similar
context.  In the third phase, selection learning, the most
probable combination of hypothesised structure is selected.

We believe the implementation of ABL presented here will be beneficial
for further explorations in applying alignment-based grammar inference
to different tasks and will turn out to be useful for unsupervised
machine learning research in NLP and other areas.

The current implementation is derived from the software used in earlier
work.  Several improvements and extensions have been made
since.  For example, the input and output formats have been simplified
and the suffix tree algorithms have been added.  The results generated
using this system should still be the same with respect to the
original system.