I am working on a project of "AUTOMATED ESSAY EVALUATION". There will be an student answer and a standard/model answer. These two answers are formatted into a standard format. Thereafter these two formats are compared. According to the matching evaluation will be done. I am wondering how to transform english sentence into the standard format like FOPL, Object oriented format, associative network, frame structure etc.
-
1My understanding is it's not possible to represent more than a limited and artificial subset of natural language in first-order predicate logic. There's a major world-wide research effort to construct formal models of individual natural languages (as well as the language faculty generally), but none of them are complete. But maybe I'm not understanding what you want to do?– Gaston ÜmlautMar 3, 2012 at 5:50
-
Hi Amit. Thanks for your question. The task you describe appears to be parsing, and fully answering it requires extensive discussion which won't fit this Q&A, and as @GastonÜmlaut said none of the available methods are perfect. Maybe this is why you're getting downvoted. I suggest you be more specific in what you are trying to do, and reword the explanation so we can understand clearly the context.– Louis RhysMar 3, 2012 at 6:18
-
1+1: This doesn't deserve a downvote! Just say that parsing is an unsolved problem, and that's the answer.– Ron MaimonMar 4, 2012 at 6:58
-
Thanks Ron Maimon and Louis....My fundamental question is to represent a sentence into a standard format. And i got the answer that its not possible...– AmitMar 4, 2012 at 12:06
-
@Amit: It is without a doubt possible, just linguists have not done it yet. I think I can do it personally, but I never sat down and did more than preliminary sketches, so I might be full of crap. The reason this is a problem is that you need to deal with identifying the merge operations, and BNF is just a little too domain unspecific so that there is a large exponential growth in node-types which makes hundreds of different nodes in a BNF description, most of which are related in nontrivial ways not captured by simple or extended BNF. But a slight extension of BNF should reduce things enough.– Ron MaimonMar 4, 2012 at 15:01
|
Show 2 more comments
1 Answer
As others have commented, there is currently no general solution to this problem but there is existing best practice.
There has been lots of research in this area. I would characterize this as straddling the line between computational linguistics, artificial intelligence, and education/testing theory. For example, a recent text on the subject is Shermis and Burnstein's Automated Essay Scoring: A Cross-disciplinary Perspective. So this question is actually incredibly broad in scope.
There are also existing systems that perform such analysis, notably ETS's Criterion system. A complete description of that system can be found in the Association for Advancement of Artificial Intelligence site. The approach is a little different than what you seem to assume in your question. It is statistical in nature; it doesn't parse the grammatical structure of the essays but rather compares characteristics of the essays to corpora of common bigrams (2-word strings) and essays with known scores. Analysis and scoring is performed using AI decision-making and voting algorithms.
answered Apr 3, 2012 at 3:06
