W 118 M

Writing a text well is hard work. One has to pay attention to many rules of various sorts. Rules about the genre of the text, about how it should look overall, correct spelling or grammar. This module within the word processor is there to help one in writing an error free and easily readable text. How the module does this and how successful it is in doing so, is what I will present in this paper.

Method

To show how grammar check works and to examine it further I ran the module on two documents. Both of them are research papers written on introductions and for the Writing and Research Skills seminar in the 1997 spring semester. For the anonymity of the authors I will refer to them as Male and Female. These texts have been already evaluated and they are the final versions.

For the tests I ran I used the Grammar Check module of Word 7.0. Since I do not have much experience with the program I set it to use apply every single rule it knows both for grammar and style (Figure 1). This makes 19 grammar rules and 26 stylistic ones. For a list of these rules see Appendix 1.

Figure 1 The settings

Results and Discussion

The program has found numerous words or phrases it considered had errors in both documents. Altogether 14 types of errors were indicated (11 types in Male’s and 14 types in Female’s) (see appendix 2). The most warnings were given about passive structures (37 times) and about very long sentences (21 times) (figure 2 and 3). Using long sentences in a text can significantly lower its readability as well as using many passive structures. These may be complicated and hard to understand for the reader, so their number should be kept low by using active structures and shorter sentences.

Figure 2 Passive structures

Figure 3 A very long sentence

The next big group of errors was considering word choice. Wordy expressions, stock phrases, phrases that the module considered too long or/and suggested a shorter version instead. These mistakes also only affect the readability of the text and they are grammatically correct as they are. These are stylistic errors (figures 4-6)

Figure 4 Informal usage

Figure 5 Nonstandard modifier

Figure 6 Stock phrase

Sentences in which stylistic errors are detected by the grammar check program are grammatically correct usually. The difference these errors mean lies in the readability of the text. As a general rule in can be stated the shorter and simpler a sentence, phrase or even a word is the easier it is to understand what is meant by it. With long clumsy sentences, many passive structures and Latin origin words a writing may seem to have been written by a more intelligent person but much less people will read it for the same reason. Therefore it is highly advisable to revise and see if the structure the module found can be simplified in any way.

Detecting stylistic errors the module rarely makes a mistake. However when it comes to real grammatical rules the flaws of the program start surfacing. Out of the 14 grammatical mistakes (not counting the pronoun errors) the module claimed to have found only 5 were real mistakes. In the rest of the cases the program failed to notice an important word, or that the word in question was a part of a phrase and other such malfunctions occurred. These mistakes of the grammar check module come from the programming flaws and, mainly, from the complexity of the language. It is impossible to create a perfect grammar check program without artificial intelligence that can tell the program what is meant by the sentence or word and what to do with it.

The four types of grammatical errors found by the module are pronoun errors, run-on sentences, clause errors and subject-verb agreement. With the pronoun errors the module had no problem since that is mainly a word choice type of problem too. In the other cases however it made several mistakes, in 2/3 of the cases in fact, making one click on the ignore icon one too often.

Figure 7 Supposed clause error

In the clause error given in figure 7 the program failed to see that the missing clause is in the previous sentence. The module has no capability to look at the sentence as a part of the whole, therefore it can make such mistakes much too often.

Figure 8 Supposed subject-verb agreement error

In figure 8 also the program failed to see that the word ‘test’ is not in the meaning the module expects to be in.

Figure 9 Supposed run-on sentence error

At the end of a check the program than gives a table with figures about word count, averages and readability statistics (figure 10). These statistics show the percentage of passive structures in the text and the reading ease scores by different methods. The percentage of passives should be under 40% and if over revision and use of active structures is recommended.

Figure 10 Text statistics

Conclusion

Despite of all of the grammar check modules flaws, it is worth using it. It can be good for revising oneself before submitting a paper of any kind. However it should not be relied upon as a final reviser. For it can not know your thoughts and correct by them. The program is best suited for home use where requirements are only basic, but it would not stand for an editor.

Appendix 1 - Rules applied

Grammatical:
Agreement with here or there
Clause errors
Commonly confused words
Double negatives
Format errors
Informal usage
Jargon words
Mass vs. Count; A vs. An
Nonstandard expressions
Nonstandard modifiers
Noun phrase consistency
Passive verb usage
Pronoun errors
Punctuation errors
Repetitive expressions
Subject-word agreement
System messages
Verbal group consistency
Word usage
Stylistic:
Check quoted text
Archaic expressions
Clichés
Contractions
Gender-specific expressions
Homonyms
Inappropriate prepositions
Informal expressions
Jargon expressions
Misspelled expressions
Misspelled foreign expressions
Misused words
Multiple negation
Open vs. Closed spelling
Overused phrases
Possible word confusions
Pretentious words
Redundant expressions
Sexist expressions
Stock phrases
Ungrammatical expressions
Vague quantifiers
Weak modifiers
Wordy expressions I
Wordy expressions II

Appendix 2 - Search results:

(Hits: number of times error type occurred altogether, Mistakes: program could not apply the rule properly [ percent of accuracy])
Male:

Hits Mistakes [Accuracy]
Confused words 4 4 [0 %]
Passive 20 - [100 %]
Long sentence 8 - [100 %]
Subject-word agreement 4 3 [25 %]
Pronoun errors 13 - [100 %]
Wordy expressions 3 - [100 %]
Clause errors 2 2 [0 %]
Informal usage 3 1 [66 %]
Pretentious words 1 - [100 %]
Weak modifier 1 - [100 %]
Redundant expression 1 - [100 %]

Female:

Hits Mistakes [Accuracy]
Passive structures 17 - [100 %]
Wordy expressions 11 2 [81 %]
Long sentences 13 - [100 %]
Run-on sentence 2 1 [50 %]
Vague quantifier 1 - [100 %]
Pretentious words 3 - [100 %]
Clause errors 2 1 [50 %]
Subject word agreement 3 2 [33 %]
Stock phrase 1 - [100 %]
Confused words 6 6 [0 %]
Weak modifier 2 - [100 %]
Redundant expression 1 - [100 %]