Natural Language Query is a context-based natural language system. Perceptive Search already knows a great deal about the text in your index. It knows the index vocabulary, word usage characteristics and word distribution. Natural Language query uses all this information, combined with a contextual rule-base, to read your Natural Language queries.
Without any configuration, Natural Language query should give correct results around 90% of the time, when asked questions that make sense within the context of the index. Asking a question "off the topic" will yield much the same result as asking a question about RAM at a farming conference - you may get a reply, but it probably will not be what you expected.
Although Natural Language Query works very well without any configuration, there are actions you can take to increase its accuracy.
When Natural Language Query fails to answer a question correctly, it tends to be for one of two reasons:
The terminology the user employed does not match what's stored in the index.
For example, the user typed What have we got on Chicago?, whereas the index never, or rarely, mentions "Chicago". It always refers to "Illinois".
Natural Language query respects the Perceptive Search synonyms, WordNet Thesaurus and tense expansion just like the other query methods.To narrow searches, users may have to turn off the synonym rings, tense expansion and thesaurus options in Perceptive Search Query preferences.
Natural Language Query incorrectly judges a word to be an important part of the query, when it is not.
Natural Language query contains sophisticated logic to parse your request, figure out which words and phrases are the most important, and how it all fits together. However, depending on your index and the mode of expression your user employs, Natural Language query may sometimes go off chasing a red herring and completely miss the intent of the question. For example, in the query, 'Tell me about blood pressure variance', if Natural Language query cannot find anything about "blood pressure variance" it might, depending on what is in your index, give you results based just on "blood pressure". Alternatively, it might incorrectly judge that "variance" is the most important part of the query and give you results either heavily or entirely biased towards that word.
You can configure Natural Language Query and give it a list of words which it might expect to encounter in queries, but to which it should assign no significance. This list works in conjunction with the Common Words lists (which are used in plain English queries), and can be thought of as an extension to the list.
These additional words (known as Natural Language Interface (NLI)) are stored in the ISYS.NLI file, which may be edited using an text editor (such as Notepad). The NLI file will also allow you to declare a word to be of low information value just in the context of Natural Language query. It does not affect whether or not the word is indexed and how it is treated outside Natural Language query. This allows the Perceptive Search index to maintain precision for Command Based queries.