How it Works
CognitionSearch™ will attempt to retrieve on all of the concepts it finds in your search sentence. It is best to avoid unnecessary concepts. Ineffectual queries result from using preambles or providing too much background information to CognitionSearch. Search sentences that combine concepts or phrases with lowercase "and" or "or" will usually only get retrievals if both concepts or phrases are in the same sentence. If you don't get any results try two separate searches or use Boolean operators (see below).
To search the database, type an ordinary English sentence. For best results:
Use grammatical English.
Use proper capitalization and capitalize acronyms.
- U.S., not u.s. or us
- Mary Smith, not MARY SMITH or mary smith.
- Tell me about the General Services Administration? not Tell me about the general services administration?
- Capitalized phrases are understood as technical terms by CognitionSearch. For example, World Financial Center is taken to be a name to be found in exactly the same form in the text. If you want to find any texts with the concepts "world," "financial," and "center" near each other, use lower case.
CognitionSearch cannot find information if it is not in the files being searched. Search sentences about topics not covered in the database provided here will probably get the reply "CognitionSearch has no information on your query."
CognitionSearch is an information retrieval system, not a question-answering system. CognitionSearch does not have the ability to make calculations or extrapolations about the text. If the search sentence asks "What constitutes failure of fiduciary duty?", CognitionSearch will not answer the question directly, but will show sections of the text about failure of fiduciary duty. If the search sentence asks "How long is the statute of limitations for a breach of contract?" CognitionSearch will retrieve information about time periods dealing with a statute of limitations and a breach of contract, if there is any such information in the text. Similarly, evaluative questions such as "What is the best defense for fraud?" will only retrieve information if the text contains explicit statements about the best fraud defenses. If searches like those above fail to produce any hits, try rephrasing them by simply stating the topic of the search, e.g., "Tell me about fraud defenses." or "failure of fiduciary duty."
CognitionSearch includes a Boolean search capability which allows users to combine English phrases or sentences with Boolean connectives to further refine a search. Seven Boolean connectives are supported, as follows (note: in the following table, Q1 and Q2 each represents a full concept query phrase):
|Q1 AND Q2||Documents in which Q1 and Q2 both matched.|
|Q1 OR Q2||Documents in which either Q1 or Q2 matched.|
|Q1 WITH Q2||Documents in which Q1 and Q2 occur in the same sentence.|
|Q1 WITHIN 20 Q2||Documents in which Q1 occurred within 20 words of Q2|
|Q1 AND NOT Q2||Documents which matched Q1 but did not match Q2|
|Q1 NOT WITH Q2||Documents which matched Q1 and no sentence that matched Q1 also matched Q2|
|Q1 NOT WITHIN 25 Q2||Documents which matched Q1 and where Q2 was not matched within 25 words of any place where Q1 matched|
(NOTE: The difference between a search using the WITH connective and a search using no Boolean connectives lies in how the search concepts are determined. In a search having the WITH connective, each subquery is interpreted independently of the other subquery. When no connectives are present, all of the words in the query provide context for determining the meaning of the other words.)
These connectives can be nested with parentheses to form full Boolean expressions.
The WITHIN connective is actually fairly flexible. By default it is expressed in terms of word positions, but it can be expressed in terms of sentences as well, and it can be abbreviated or expanded in various ways. For example:
|W/IN20||within 20 words|
|W/IN5W||within 5 words|
|W/IN2S||within 2 sentences|
|WITHIN 20 SENTENCES OF||within 20 sentences of...|
|WITHIN 20 WORDS OF||within 20 words of|
|WITHIN 20 WORDS||within 20 words of|
|WITHIN 20 SENTENCES||within 20 sentences of|
Note that all the letters in the connective must be capitalized and there may not be more than one space character separating the words.
These connectives can be nested with parentheses to form full Boolean expressions.
If part of a query is quoted, the engine will only report exact matches for the quoted part of the query. In other words, it will not show documents where a search term from the quoted part of the query was matched against a term with a similar meaning. For example, the query
might retrieve documents that refer to specific types of tree frogs, whereas
would only return documents that actually have the word "tree" in them (although since "frog" is not in quotes, the documents might not have the literal word "frog" in them). Note, however, that quoted terms are disambiguated. In other words, in the above example, "tree" would only retrieve exact matches that refer to the type of plant and would not retrieve instances of the string that refer to abstract representations such as a family tree. If you wish to perform a string search (in which terms are not disambiguated), see the Pattern Search section below, or use single quotes, as in
which will return any documents that contain the string "tree," regardless of its meaning (still providing that it's in the same sentence as the concept "frog"). Note that, unlike in pattern search below, you may include more than one word inside the quotation marks, although this does not imply a requirement that the words be adjacent in the document base.
The above quote types (single- and double-quoting) result in exact matches, but there are other types that don't. For example, you could search for
and normally in that context you would only get references to work stoppages, but with the curly-bracket quotes, you would retrieve to every meaning of "strike", not just the work stoppage one. Not only that, it would retrieve to every synonym of every one of those meanings.
If you quote a group of more than one word, as in, for example
the query will only match the words inside the quotes if they are next to each other and in the order given, so the above would not match "frog tree" or "trees with leaves that are green like a frog". If you want to match a set of words in a particular order in this way, but do not want to require an exact match, you can use square-bracket quotes:
The above will match "scarlet carnation" and "rufous rose", but not "flowers that look red".
Pattern terms are not matched against indexed concepts, but rather against the (stemmed) strings stored in the index for each document in a project. For example, the pattern term
would match any instance of the exact string "dog" in a document.
There are a number of different pattern types. The following sections will lay out each type, with examples.
The Raw Text Pattern
which start with a forward slash, continue with a word, and end with a forward slash, match the given word exactly. If the pattern were
it would match that exact spelling and capitalization pattern, i.e. "dOg", but not "dog". Note, however, that the strings in the index are stemmed, so the pattern /swims/ would be unlikely to match anything, whereas /swim/ would
The Case-Insensitive Flag
Starting from the raw-text pattern, one may add flags to create different behavior. One such flag is "i," the case-insensitive flag. Adding this flag indicates that the pattern should match the word given, regardless of case differences. Thus
would match “dog,” “dOg,” “DOG,” “Dog,” etc.
It is possible also to search using the wildcards "*" and "?." The wildcard "*" indicates an indefinite length of unspecified characters, whereas "?" indicates one specified character. A wildcard search pattern must be between angle brackets rather than slant brackets. For example,
will match any word that begins with “d” and ends with “g,” whereas
will match any three-letter word that begins with "d" and ends with "g." When searching for names like "Liebowitz," one might use the wildcard pattern
and thus get "Liebowitz," "Leibowitz," "Liebowicz," etc.
Case-Insensitive Wildcard Search
The "i" flag (currently no others) may be added to a wildcard search pattern to make it case-insensitive. Thus
will match "dog" and "dig," but not much else, whereas
will also match “Dog,” “Dig,” etc.
Just as you can quote a group of words, as in "tree frog" and require them to appear next to each other in a particular order, you can also place multiple words inside a pattern match and require the same. Thus,
matches exactly the words "red" and "flower", together and in that order. It would not match "flower red" or "the red walls of his house were covered with flowers.", although the query
would match them.
Required and Optional Terms
One may mark a term or pattern as especially significant by preceding it with a plus sign (+). In this case no results will be returned that do not contain the marked term. For example, in the query
tree frogs of Asia
it might be the case that no documents match all three search terms, but some documents match tree+Asia and some match frog+Asia. Normally, in that case the search engine will offer the choice of those two breakdowns. However, if the query had been
tree +frogs of Asia
only the frog+Asia results would have been returned. In a quoted query, the plus sign goes inside the quotes:
Similarly, one may mark a term or pattern as optional by preceding it with a minus sign (-). In this case no document will be disqualified on the basis of not containing the optional term, although the optional term will affect sense selection and documents containing the term will be preferred in sorting over documents not containing it. For example, in the query
strike a -match
the word "match" will indicate that the desired meaning of "strike" is the one associated with striking a match, but documents returned do not necessarily need to actually contain the word "match."