“The limits to computer-based grammar checking for
foreign language learners of English?”
Philip
Bolt
Department of English, Hong Kong Polytechnic
University
egphbolt@hkpucc.edu.hk
Although much work has been carried out in the areas related to computer-based
grammar checking programs (those wh ich specifically deal with syntactically
ill-formed input), the history of this endeavour is not an especially successful
one for either native speaker language or the possibly less complex language of
the English as a foreign language (EFL) learner.
A number of factors militate against the development of a program about which we
can be confident in its capacity to identify faulty syntax and provide
linguistically appropriate reformulations. Yet, with sufficiently fine-grained
lexical and syntactic data and sensitive processes (including those involving
the learner to some degree), it is possible in some cases to identify the
syntactic aspects of a faulty sentence and to say, with reasonable confidence,
what is wrong. In many cases, it is possible to locate a problematic area,
rather than a single feature and to suggest one or more reformulations involving
that area, and possibly beyond. In other cases, however, the lexical and
structural complexities, the range of problems within and across phrase and
clause edges and the range of potential solutions make structural analysis, and
problem identification and reformulation, much more difficult. This suggests
that there might be a threshold (or rather series of different thresholds) of
complexity/corrigibility beyond which a grammar checker might not be
effective.
ISCA (Interactive Sentence Constructor and Analyser) is a program which provides
the framework for the demonstration and discussion of what is involved in the
design, construction and operation of a grammar checking program for EFL
students, together with the features which collectively contribute to the above
thresholds. In addition to the graphical depiction of the program and related
issues provided by the poster, likely discussion will be approached through the
demonstration of a number of well and ill-formed sentences, which illustrate the
data and process involved in sentence construction, correction and the user's
contribution to these. It is suggested that these sentences illustrate a range
of structural features - including the number, nature and distribution of items
- word-form/word-class pairings - the directness of such pairings to phrases and
clause elements, the availability of a range of key words, and the range and
type of possible reformulations - which contribute to the notion of a number of
complexity/corrigibility thresholds. The use of the software will provide a
`hands-on' opportunity and complement the information provided on the poster.
Together, they will give those interested an opportunity for observation and
discussion of the program and its underlying philosophy and ontology.
The presentation might appeal to a) those disdainful of the performance of
commercial grammar (style) checking programs, but who may not have fully
appreciated what is involved, b) those who have (or are) working on parsers and
grammar checking programs of their own, and, c) those who feel that syntax is
not important for writers and readers.