Abstract:

In order to leave more time for EFL teachers to work on higher-level re-writing tasks, we decided to develop a computer grammar checker. The first stage of development was devoted to error analysis of 125 writing samples collected from our students. We found 1659 errors and classified them into 14 main types and 93 subtypes. The analysis served as the basis for constructing a taxonomy of mistakes and ranking the categories according to frequency of occurrence and comprehensibility. To implement the grammar checker, we first built a small electronic dictionary with 1402 word stems and necessary features, and designed a suffix processor to accommodate morpho-syntactic variants of each word stem. We then constructed an ATN parser, equipped with phrase structure rules and error patterns. In addition, a set of disambiguating rules for multiple word categories was designed to eliminate unlikely categories and thus increase the parser's efficiency. The current implementation detects seven types of errors and provides corresponding feedback messages. Future research will be focused on detecting more types of mistakes with greater precision and on providing appropriate editing strategies.

Author: Hsien-Chin Liou