Recursive Top-down Fuzzy Match : New Perspectives on Memory-based Parsing

With treebanks becoming available for more and more languages, their usage for the development of natural language parser has become a topical issue in NLP. This paper tries to give a new spin to this this stream of research, proposing a new direction in corpus-based parsing. Contrary to competitive approaches, this approach does not involve a chart parser which reassembles phrases extracted form a treebank. Instead, parsing proceeds via the extraction of example trees from the treebank using fuzzy pattern matching techniques. A set of adaptation rules modify the extracted example trees so as to produce the best possible parse given the current set of examples.