POSABL: ABL-based grammar induction on POS tags
Peyman Passban, Mahmoud Shokrollahi-Far · 2014
Grammar Induction (GI) is the problem of extracting hidden regularities and syntactic patterns in languages. Not only the manner of extraction is intricate but also the definition of meaningful patterns is a challenge. Alignment Based Learning (ABL) is one of the research endeavors targeting such challenges in GI. Our present research on applying ABL to POS sequences in English, Persian and Arabic languages has led us to POSABL yielding the enhanced recall of 83%, 74% and 100% for the three languages, respectively. Our experimental results on minimizing the size of the tag-set and the maximum size of train data required for our GI are considerable, as well. This research trend on POS sequences is hoped to pave the way more for GI on different languages of creation, say DNA sequences.