BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Finite-State Transducers as a Theory of Dependency Structured Natu
 ral Language - Anssi Yli-Jyrä\, University of Helsinki
DTSTART:20170922T110000Z
DTEND:20170922T120000Z
UID:TALK81861@talks.cam.ac.uk
CONTACT:Anita Verő
DESCRIPTION:Important families of dependency graphs are STRONGLY equivalen
 t to context-free\nlanguages whose strings encode the graphs.  The consequ
 ences of this observation\n(in ACL 2017) increase (i) the scope of the cla
 ssical context-free grammars by\nallowing them to generate non-trees\, (ii
 ) the relevance of strictly locally\ntestable languages to nonlocal graphi
 cal properties\, and (iii) the applicability\nof superbracketing schemes b
 eyond the bracketing of single-headed\nphrase-structures.  The second part
  of the talk is about the work in progress\nand considers minimal recursio
 n in the encoded graphs.  My preliminary\nexperimental results (in FSMNLP 
 2017) on the Universal Dependencies v2 treebanks\nindicate that a quite sm
 all subset approximation of an encoded family of\ndependency graphs would 
 already capture 99.994% of the primary dependency\nstructures if the encod
 ing scheme is IMPLICITLY extended to crossing brackets.\nThis raises inter
 esting questions about the expressiveness of the finite-state\ngrammar ove
 r graph banks and its potential applications to linear time encoding\nand 
 decoding tasks in dependency parsing and machine translation.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR