Sinhala and frequent Code‐Mixing involving a few English expressions. Just like
Sarfraz in her study Abeywickrama also followed the guidelines of Ellis 1995. A
sample of written works were collected from 60 students who are in the first and
the second academic year of their Degree programmes. They were provided with
the topics “An Unforgettable Day in Your Life” and “My University Life” and were
asked to write on it in 200 to 250 words. They were given sufficient time to write
Ellis, 1997 starting with an outline, then a first draft and a final one.
The findings of his study showed that the highly objective and outcome
oriented investigation reflects negative first language transferinterference is not
the major cause for errors in the English composition that were written by
Sinhala speaking undergraduate students.
4. Joel R. Tetreault and Martin Chodorow’s study 2008
In their study entiltled: The Ups and Downs of Preposition Error Detection in
ESL WritingTetreaultand Chodorow describe a methodologyfor detecting
preposition errors in the writingof non‐native English speakers. They were
interested to conduct that research due to the fact that non‐native English
writers are great in number. Those people often made errors in using
prepositions. The objective of the research was to find out how the ups and
downs of preposition error happen in their writing production.
The methodology used in the research was the one described in Chodorow
and Leacock, 2000 for the task of evaluating the usage of nouns, verbs and
30
adjectives. The central idea is to skew the annotation corpus so that it contains a
greater proportion of errors. They then took the following steps in the
procedure by processing, dividing, combining the samples into an annotation
set, judging, calculating and using the values to calculate precision.
The system performs at 84 precision andclose to 19 recall on a large set of
studentessays. In addition, they address theproblem of annotation and
evaluation inthis domain by showing how current approachesof using only one
rater can skewsystem evaluation. They present a samplingapproach to
circumvent some of the issuesthat complicate evaluation of error
detectionsystems. This
paper has two contributions to the field of error detection in non‐ native writing.
First, it discussed a system that detects preposition errors with high precision
up to 84 and is competitive with other leading methods. It used an ME
approach augmented with combination featuresand a series of thresholds. This
system is currently incorporated in the Criterion writing evaluation service. Second,
it showed that the standard approach to evaluating NLP error detection systems
can greatly skew system results when the annotation is done by only one
rater. However, one reason why a single rater is commonly used is that building
a corpus of learner errors can be extremely costly and time consuming. This
makes using multiple raters possible since less time is required to assess the system’s
performance.
5. Dominika Uhrikova’s study 2011
