Implementing Alignment-Based Learning

In this article, the current implementation of the AlignmentBased Learning (ABL) framework (van Zaanen, 2002) will be described. ABL is an unsupervised grammar induction system that is based on Harris’s (1951) idea of substitutability. Instances of the framework can be applied to an untagged, unstructured corpus of natural language sentences, resulting in a labelled, bracketed version of that corpus. Firstly, the framework aligns all sentences in the corpus in pairs, resulting in a partition of the sentences consisting of parts of the sentences that are equal in both sentences and parts that are unequal. Since substituting one unequal part for the other results in another valid sentence, the unequal parts of the sentences are considered to be possible (possibly overlapping) constituents. Secondly, of all possible constituents found by the first phase, the best are selected.