BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learnable representations for natural language - Alexander Clark -
  Royal Holloway University of London
DTSTART:20100205T120000Z
DTEND:20100205T130000Z
UID:TALK22724@talks.cam.ac.uk
CONTACT:Laura Rimell
DESCRIPTION:The Chomsky hierarchy was explicitly intended to represent the
 \nhypotheses output by distributional learning algorithms\; yet these\nsta
 ndard representations are well known to be hard to learn in an\nunsupervis
 ed fashion\, even under quite benign learning paradigms\,\nbecause of the 
 computationally complexity of inferring rich hidden\nstructures like trees
 . Nonetheless there is much interest in NLP  on\nunsupervised learning of 
 syntax: trying to infer gold standard trees\nfrom unannotated data.\nBut t
 his approach is misguided -- we don't know what the right\nrepresentations
  are but we do know they are learnable\, since children\ndo in fact acquir
 e such representations. Accordingly\, we  explore a\ndifferent approach --
  building representations or grammars that are\nintrinsically learnable.\n
 \nThis research direction  involves abandoning the standard models and\nde
 signing new representation classes for formal languages that are\nrichly s
 tructured but where the structure is not hidden but based on\nobservable s
 tructures of the language.\nWe illustrate this approach by looking at algo
 rithms for learning\nregular languages using deterministic automata\, and 
 then move on to\nalgorithms for learning context free and context sensitiv
 e languages.\nThe largest and most powerful class\, based on the theory of
  residuated\nlattices\, may be rich enough to represent natural language s
 yntax\;\nthese grammars are cubic time parseable and are efficiently learn
 able.\nThe class of languages defined by these representations contains al
 l\nregular languages\, many but not all context free languages\, and some\
 ncontext sensitive languages\; it thus seems a plausible candidate for\nth
 e class of possible natural languages.
LOCATION:SW01\, Computer Laboratory
END:VEVENT
END:VCALENDAR