In order for an expert system to be an effective tool, people must be able to interact with it easily. To facilitate this interaction, the expert system must have the following two capabilities in addition to the ability to perform its underlying task:

Explain its reasoning. In many of the domains in which expert systems operate, people will not accept results unless they have been convinced of the accuracy of the reasoning process that produced those results. This is particularly true, for example, in medicine, where a doctor must accept ultimate responsibility for a diagnosis, even if that diagnosis was arrived at with considerable help from a program. Thus it is important that the reasoning process used in such programs proceed in understandable steps and that enough meta-knowledge (knowledge about the reasoning process) be available so the explanations of those steps can be generated.

Acquire new knowledge and modifications of old knowledge. Since expert systems derive their power from the richness of the knowledge bases they exploit, it is extremely important that those knowledge bases be as complete and as accurate as possible. But often there exists no standard codification of that knowledge; rather it exists only inside the heads of human experts. One way to get this knowledge into a program is through interaction with the human expert. Another way is to have the program learn expert behaviour from raw data.

One could imagine a naive spell checker as a large corpus of correct words. Thus if a word in the text being corrected does not match with one in the corpus then it results in a spelling error. An exhaustive corpus would of course be a mandatory requirement.

Spell checking techniques can be broadly classified into three categories:

i. Non-Word Error Detection:

This process involves the detection of misspelled words or non-words. For example -

The word soper is a non-word, its correct form being super (or maybe sober).

The most commonly used techniques to detect such errors are the N-gram analysis and Dictionary look-up. N-gram techniques make use of the probabilities of occurrence of N-grams in a large corpus of text to decide on the error in the word. Those strings that contain highly infrequent sequences are treated as cases of spelling errors. Note that in the context of spell checkers we take N-grams to be a sequence of letters (alphabet) rather than words. Here we try to predict the next letter (alphabet) rather than the next word. These techniques have often been used in text (handwritten or printed) recognition systems which are processed by an Optical Character Recognition (OCR) system.

The OCR uses features of each character such as the curves and the loops made by them to identify the character. Quite often these OCR methods lead to errors. The number0, the alphabet O and D are quite often sources of errors as they look alike. This calls for a spell checker that can post-process the OCR output. One common N-gram approach uses tables to predict whether a sequence of characters does exist within a corpora and then flags an error. Dictionary look-up involves the use of an efficient dictionary lookup coupled with pattern-matching algorithms (such as hashing techniques, finite state automata, etc.), dictionary partitioning schemes and morphological processing methods.

ii. Isolated-Word Error Correction:

This process focuses on the correction of an isolated non-word by finding its nearest and meaningful word and makes an attempt to rectify the error. It thus transforms the word soper into super by some means but without looking into the context.

This correction is usually performed as a context independent suggestion generation exercise, the techniques employed herein include the minimum edit distance techniques, similarity key techniques, rule-based methods, N-gram, probabilistic and neural network based techniques.

Isolated-word error correction may be looked upon as a combination of three sub-problems: Error detection, Candidate (Correct word} generation and Ranking of the correct candidates. Error detection as already mentioned could use either of the dictionary or the N-gram approaches. The possible correct candidates are found using a dictionary or by looking-up a pre-processed database of correct N-grams. Ranking of these candidates is done by measuring the lexical or similarity distance between the misspelled word and the candidate.

iii. Context dependent Error detection and correction:

These processes try, in addition to detect errors, try to find whether the corrected word fits into the context of the sentence. These are naturally more complex to implement and require more resources than the previous method. How would you correct the wise words of Lord Buddha -

"Peace comes from within"

if it were typed as -

"Piece comes from within" ?

Note that the first word in both these statements is a correct word.

This involves correction of real-word errors or those that result in another valid word. Non-word errors that have more than one potential correction also fall in this category. The strategies commonly used find their basis on traditional and statistical natural language processing techniques.

Refer 2^nd Question & Answer OR CLICK HERE

There are four ways of handling sentences such as these:

9.1 All Paths:

Follow all possible paths and build all the possible intermediate components. Many of the components will later be ignored because the other inputs required to use them will not appear. For example, if the auxiliary verb interpretation of "have" in the previous example is built, it will be discarded if no participle, such as "taken," ever appears. The major disadvantage of this approach is that, because it results in many spurious constituents being built and many dead end paths being followed, it can be very inefficient.

9.2 Best Path with Backtracking:

Follow only one path at a time, but record, at every choice point, the information that is necessary to make another choice if the chosen path fails to lead to a complete interpretation of the sentence. In this example, if the auxiliary verb interpretation of "have" were chosen first and the end of the sentence appeared with no main verb having been seen, the under stander would detect failure and backtrack to try some other path. There are two important drawbacks to this approach. The first is that a good deal of time may be wasted saving state descriptions at each choice point, even though backtracking will occur to only a few of those points. The second is that often the same constituent may be analysed many times. In our example, if the wrong interpretation is selected for the word "have," it will not he detected until after the phrase "the students who missed the exam" has been recognized. Once the error is detected, a simple backtracking mechanism will undo everything that was done after the incorrect interpretation of "have" was chosen, and the noun phrase will be reinterpreted (identically) after the second interpretation of "have" has been selected. This problem can be avoided using some form of dependency-directed backtracking, but then the implementation of the parser is more complex.

9.3 Best Path with Patch up:

Follow only one path at a time, but when an error is detected, explicitly shuffle around the components that have already been formed. Again, using the same example, if the auxiliary verb interpretation of "have" were chosen first, then the noun phrase "the students who missed the exam" would be interpreted and recorded as the subject of the sentence. If the word "taker" appears next, this path can simply be continued. But if "take" occurs next, the under stander can simply shift components into different slots. "Have" becomes the main verb. The noun phrase that was marked as the subject of the sentence becomes the subject of the embedded sentence "The students who missed the exam take it today." And the subject of the main sentence can be filled in as "you," the default subject for imperative sentences. This approach is usually more efficient than the previous two techniques. Its major disadvantage is that it requires interactions among the rules of the grammar to be made explicit in the rules for moving components from one place to another. The interpreter often becomes ad hoc, rather than being simple and driven exclusively from the grammar.

9.4 Wait and See:

Follow only one path, but rather than making decisions about the function of each component as it is encountered, procrastinate the decision until enough information is available to make the decision correctly. Using this approach, when the word "have" of our example is encountered, it would be recorded as some kind of verb whose function is, as yet, unknown. The following noun phrase would then be mterpreted and recorded simply as a noun phrase. Then, when the next word is encountered, a decision can be made about how all the constituents encountered so far should be combined. Although several parsers have used some form of wait-and-see strategy, one, PARSIFAL [Marcus, 1980], relies on it exclusively. It uses a small, fixed-size buffer in which constituents can be stored until their purpose can be decided upon. This approach is very efficient, but it does have the drawback that if the amount of look ahead that is necessary is greater than the size of the buffer, then the interpreter will fail. But the sentences on which it fails are exactly those on which people have trouble, apparently because they choose one interpretation, which proves to be Wrong.

Refer 3^rd Question & Answer OR CLICK HERE