I’m currently reading ‘Artificial Intelligence in Second Language Learning: Raising Error Awareness‘ written by Marina Dodigovic and published in November 20, 2005.
Below is an excerpt from page 107:
“Commercially available grammar checkers indeed have another weakness. They cannot evaluate the semantics or the full syntax of the given text and as a result allow text like the following:
Their are know miss steaks in my pepper be cause my word processor includes soft wear witch checks my spelling. The problem is that it doesn?t correct errors in punctuate and it will not fined words that have bean miss used but that are spelled write. An if I write badly constructed sentences it won?t correct them four me. (Sanders, 2000)
The reason why the latest version of MS Word grammar check program has not noticed that ‘punctuate’ in the second sentence should be a noun, ‘punctuation’, rather than a verb, as evident from the text, is because it most likely looks at two or three adjacent words at a time and calculates the statistical probability for their simultaneous co-occurence in a text. It does not look at the sentence as a whole. This kind of parser is called a probabilistic parser.”
Dr. Dodigovic is probably talking about MS Word 2003 and not being able to resist my curiosity I have tried the example used by her on the latest version of MS Word 2007.
It seems like the situation got a little bit better but not that much. MS Word 2007 still can’t detect the ‘punctuate’ error and even though it provides some good feedback about some errors it is still far from perfect, at least in terms of being able to help non-native speakers of English. So the complaints of Dodigovic are still very relevant.
I haven’t tried the example using OpenOffice.org Write thus I don’t know if it can perform any better. Nevertheless, the current situation is yet another argument that statistical language processing, even though very useful for lots of practical purposes is not a silver bullet for processing natural language at the level of an ordinary human competence.