Talk by Mark Daibhidh Anderson: Increasing NLP parsing efficiency with chunking

Abstract:
The goal of the Fastparse project is to develop NLP parsers with
speeds suitable for large-scale analyses without forsaking accuracy.
A ‘Chunk-and-pass’ parsing technique influenced by a psycholingusitic
model, where linguistic information is processed not word-by-word
but rather in larger chunks of words, will be introduced. This
entails chunking linguistic data and subsequently parsing the components
inside the chunks with a simple and fast parser and the abstract
representation of the sentence constructed with chunks to a more
robust but slower parser. Preliminary results will be presented
which show that it is feasible to compress linguistic data into
chunks without significantly diminishing parsing performance and
potentially increasing the speed.

Bio:
Mark started his research career in the Nuclear and Hadron
Physics Research Group at the University of Glasgow. He then changed
research path and undertook an MSc in Informatics at the University
of Edinburgh. Latterly he has begun a PhD at the University of
Coru~na in the Fastparse group where he works on developing more
efficient NLP parsers.