University of Cambridge > Talks.cam > NLIP Seminar Series > Learning hard chart constraints for efficient context-free parsing

Learning hard chart constraints for efficient context-free parsing

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Thomas Lippincott.

In this talk, I’ll present some recent work in learning hard constraints for cells within a context-free parsing chart, to reduce parsing time. Each cell in the chart represents one of the O(n^2) substrings of the input string, and characteristics of each substring can be used to decide how much work to do in the associated chart cell. I’ll discuss finite-state models for tagging chart constraints on words, including methods for bounding the worst-case complexity of the parsing pipeline to quadratic or sub-quadratic in the length of the string. Empirical results will be presented for English and Chinese, achieved by constraining various high accuracy parsers. Finally, I will present a generalization of these finite-state approaches that performs a quadratic number of classifications (one for each substring) to produce further (finer) constraints on the amount of processing within each cell. This latter approach has the nice property of being trained on maximum likelihood parses, rather than reference parses, making for a straightforward method for tuning parsing efficiency to new tasks and domains.

This talk is part of the NLIP Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2019 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity