University of Cambridge > Talks.cam > NLIP Seminar Series > Finite-State Transducers as a Theory of Dependency Structured Natural Language

Finite-State Transducers as a Theory of Dependency Structured Natural Language

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

If you have a question about this talk, please contact Anita Verő.

Important families of dependency graphs are STRONGLY equivalent to context-free languages whose strings encode the graphs. The consequences of this observation (in ACL 2017 ) increase (i) the scope of the classical context-free grammars by allowing them to generate non-trees, (ii) the relevance of strictly locally testable languages to nonlocal graphical properties, and (iii) the applicability of superbracketing schemes beyond the bracketing of single-headed phrase-structures. The second part of the talk is about the work in progress and considers minimal recursion in the encoded graphs. My preliminary experimental results (in FSMNLP 2017 ) on the Universal Dependencies v2 treebanks indicate that a quite small subset approximation of an encoded family of dependency graphs would already capture 99.994% of the primary dependency structures if the encoding scheme is IMPLICITLY extended to crossing brackets. This raises interesting questions about the expressiveness of the finite-state grammar over graph banks and its potential applications to linear time encoding and decoding tasks in dependency parsing and machine translation.

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.

 

© 2006-2017 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity