Primarily motivated by business and office requirements, the design of CorrecText resembles more that of a grammatical checker than that of a stylistic processor. The system detects mostly non-stylistic problems, none of which were considered a priority in the design of STASEL:
• spelling errors and typos;
• grammatical problems including verb form errors, subject/verb and number disagreement, and sentence fragments;
• punctuation mistakes;
• errors in typographic conventions such as capitalization and the use of numbers.
Its stylistic component covers word usage standards and context-free writing practices of the kind found in the appendix of most writing textbooks: split infinitives, double negatives, colloquialisms, informal habits, wordy phrases and redundancies. In this sense, CorrecText functions much like STASEL, but it bases its analysis on an unsystematic and informal treatment of style. Moreover, it has no sense of goal-directed stylistic analysis and shows no teaching disposition. It is solely aimed at relieving business writers from the task of refining documents by getting rid of spelling errors, common grammatical mistakes, and prescriptive stylistic misusages.
The interest of the program lies in its use of a parsing component that
yields fewer incorrect parses and makes more complex sentence analyses than pattern-matching approaches of the Writer's Workbench sort. Since the techniques used in the parser's design are proprietary, only the account provided by Dobrin (Dob90) can shed light on the system’s operation and performance. An inspection of the examples that Dobrin fed to system reveals important portant flaws in the parsing mechanism, which ultimately make it unsuitable for language instruction purposes. CorrecText uses a probabilistic parser (referred to as a full parser by Dobrin) that does minimal semantic analysis and has no backtracking capabilities. It seemingly resolves syntactic ambiguities during the parsing process by applying stochastic clues derived from "an analysis of how errors actually occur". In dictating decisions to the parser, these clues are meant to increase the probability of catching grammatical errors, but, at the same time, this use seriously threatens the correctness of the resulting parse. In fact, Dobrin reports several instances of incorrect syntactic analyses. For instance, the following example sentence (p.69, example 2)
The Greek Islands form a barrier to progress through the Aegean sea.
is wrongly diagnosed by CorrecText as being an incomplete sentence (fragment).
Frequent incorrect parses disqualify CorrecText from any potential language instruction application, as such errors cannot be tolerated in a teaching environment. The use of a non-stochastic parser, such as the one implemented in STASEL, is essential to ensure a high degree of precision in the syntactic analyses, which, in turn, will contribute to a more effective and adequate instruction. Dobrin acknowledges this limitation when he comments on the usefulness of CorrecText's probabilistic parser:
A full parser cannot be made into a [non-stochastic] parser. The methods that allow a program to uncover accurately the maximum number of grammatical errors are not necessarily those that are useful in second-language training. (p. 71)