Automatic Mapping Clinical Notes to Medical - RMIT University

Efficient Combinatory Categorial Grammar Parsing
Bojan Djordjevic and James R. Curran
School of Information Technologies
University of Sydney
NSW 2006, Australia
{bojan,james}@it.usyd.edu.au
Abstract
Unfortunately, parsing is very inefficient because
of the large degree of ambiguity present
Efficient wide-coverage parsing is integral in natural language. This is particularly true for
to large-scale NLP applications. Unfortu- wide-coverage grammars in linguistically expresnately,
parsers for linguistically motivated sive formalisms, especially those automatically
formalisms, e.g. HPSG and TAG, are often extracted from a treebank.
too inefficient for these applications.
Many NLP systems use shallow parsing be-
This paper describes two modifications cause full parsing is too slow (Grishman, 1997).
to the standard CKY chart parsing algo- To improve the approximate structure identified
rithm used in the Clark and Curran (2006) by shallow parsers, many systems use domain-
Combinatory Categorial Grammar (CCG) specific knowledge to extract dependencies (Gr-
parser. The first modification extends the ishman, 1997; Cole et al., 1997). Ciravegna et al.
tight integration of the supertagger and (1997) show that the accuracy can be improved by
parser, so that individual supertags can be using a better parser. The ability of NLP systems to
added to the chart, which is then repaired extract useful and correct information could there-
rather than rebuilt. The second modificafore be improved substantially if the speed of full
tion adds constraints to the chart that re- parsing was acceptable.
strict which constituents can combine.
The C&C CCG parser (Clark and Curran, 2006)
Parsing speed is improved by 30–35%
without a significant accuracy penalty and
a small increase in coverage when both of
these modifications are used.
is the fastest linguistically motivated parser in the
literature, but it is still limited to about 25 sentences
per second on commodity hardware.
This paper describes two modifications to the
C&C parser that significantly improve parsing ef-
1 Introduction
ficiency without reducing accuracy or coverage.
The first involves chart repair, where the CKY
Parsing is the process of determining the syntactic chart is repaired when new categories are added,
structure of a sentence. It is an integral part of the instead of rebuilt from scratch. This allows an
deep semantic analysis that any sophisticated Nat- even tighter integration of the parser and supertagural
Language Processing (NLP) system, such as ger (described below) which results in an 11%
Question Answering and Information Extraction speed improvement over the original parser.
systems, must perform.
The second modification involves parsing with
The sentences Bob killed Alice, Alice was killed by constraints, that is, requiring certain spans to be
Bob and Bob was the man who killed Alice convey constituents. This reduces the search space con-
the same information. If we treat the sentence as siderably by eliminating a large number of con-
a bag or sequence of words by assuming limited stituents that cross the boundary of these spans.
structure, the sentences appear to be very differ- The best set of constraints results in a 10% iment.
These examples demonstrate that full parsing provement over the original parser. These con-
is necessary for accurate semantic interpretation. straints are also useful for other tasks. Finally,
Further, sophisticated linguistic analysis capable when both chart repair and constraints are used,
of modelling a wider range of phenomena should a 30-35% speed improvement is achieved while
give us the most information.
coverage increases and the accuracy is unchanged.
Proceedings of the 2006 Australasian Language Technology Workshop (ALTW2006), pages 1–8.
1