Preliminary Timeline for Honours

July 11, 2011

To better prepare for Honours, I've begun a loose sketch of my remaining timeline and the tasks I have yet to accomplish. I also set up this aesthetic countdown clock to scare the hell out of me and the other Honours students ;)

Abstract

The C&C parser currently uses the chart-based CKY algorithm. By replacing this with the shift-reduce algorithm, it is possible to get both higher speeds and even higher accuracy in parsing. The primary task will be modifying the existing C&C parser minimally so that close comparisons can be made to the existing C&C parser. Once this is accomplished, allowing ambiguity in PoS and CCG tags may allow higher accuracy through more accurate language models. Finally, through aggressive pruning of the search space, parsing speeds can be improved hopefully without a significant impact on accuracy.

Implement Shift-Reduce Algorithm

July 4th - July 24th (~3 weeks)

• Explore the C&C Parser and identify all places that need to be modified
• Modify parser to use shift-reduce algorithm instead of the CKY algorithm
• Create all data and run evaluation tests on the current C&C parser

Integrate with and Extend the C&C Parser

July 25th - 4th September (~6 weeks)

• Train perceptron/MIRA to drive search decisions during parsing
• Use beam search on top of supertagging to improve parsing speed

Optimisation and Evaluation

5th September - 2nd October (~4 weeks)

ALTA 2011 Deadline: 14th September

• In-place PoS/CCG tagging to improve accuracy and allow better disambiguation
• Investigate issues related to speed, such as aggressive pruning of search tree or engineering decisions and choices made by the C&C framework
• Evaluate against CCGbank dependencies and identify errors that occur
• Compare this parser against C&C and Zhang and Clark's shift-reduce CCG parser directly

Poster and Final Thesis Write-up

3rd October - ... (~4 weeks)

ACL 2012 Deadline: 15th January