NLP++

[Note: This topic is slated for removal. Refer also to the NLP++ section.]

NLP++™ is a novel programming language developed by Text Analysis International, Inc. It enables a developer to focus on the conceptual task of building a text analyzer while minimizing the need to delve into compiled programming languages. NLP++ combines a C++ -like syntax (called the code part) with a rule syntax (described in the Pass Files topic). The current topic is concerned primarily with the code part.

One use of NLP++ is dynamic semantics, that is, to create and use information "on the fly". Another is to embed heuristics and algorithms into the text analyzer under construction.

Regions

The NLP++ language can currently appear in the @CHECK, @CODE and @POST regions of a pass file. In a @CHECK region, the functions succeed and fail can be used to accept or reject a rule match, respectively.

The @PRE region, which operates during the pattern matching of a rule, does not yet allow NLP++ constructs.

WARNINGS:
[1] The developer must keep in mind that actions that modify the parse tree, such as single(), will disable references to the nodes of a phrase and otherwise affect the execution of NLP++ code. It is recommended that tree-modifying actions follow references to node variables.
[2] If the developer must reference a newly created node, e.g., by the group action, he must manually figure out the new element number of the created node. Note that following an action such as group, element numbers to the right of the grouped node will also change.

Tokens

See the Pass File section on Tokens.

Variables

NLP++ currently supports global variables and variables associated with particular nodes of the parse tree. The syntax for these variables is as follows:

VARIABLE	DESCRIPTION
G(varname)	Global variable.
S(varname)	Variable belonging to the suggested concept of a rule.
X(varname [,num])	Variable belonging to the numth context node starting at the root of the parse tree. Usually used to refer to the numth node of the @PATH select list. With @NODES, the preferred form is X(varname).
N(varname [,num])	Variable belonging to a node that matched the numth element of a rule phrase.

The shorthand notations N(varname) and X(varname) refer to the last element of the phrase and the last context node, respectively. The former is convenient for rules that have only one element in their phrase. The latter is convenient when using the @NODES selector. N(varname, 0) and X(varname, 0) have the identical meaning.

There is no notion of a local variable or pointer, at this time. S variables are often used as an idiom for local variables (the difference is that they "hang around" if a new node is suggested by the current rule).

NOTE: Variable names must be quoted. E.g., N("unknown word", 2)

WARNING: X(varname,num) does NOT refer to the numth member of a @NODES list. That list is not a path in the parse tree. We recommend using X(varname) with @NODES, to avoid "blindly" traversing down the parse tree. Using X(varname,2), for example, implies that you know what the nodes immediately below the root are. Such a use should be clearly documented.

Special Variable Names

To get useful information from the parse tree and other sources when analyzing a text, some special variable names have been provided. All start with a dollar sign ($). No such variable names are usable with suggested (S) variables, at present.

VARIABLE NAME	VARIABLE TYPES	DESCRIPTION
$treetext	N, X	Fetch the 'cleaned up' (removing leading and trailing, convert separators to single space) text gleaned from a node's subtree. Uses the parse tree as input, not text buffer. Analogous to $text. Difference between $treetext and $text is $text returns text from the input text buffer while $treetext returns text from the parse tree. So, if you have removed some nodes from a tree with an excise(), $treetext returns the text minus the excised elements, while $text returns everything in the input text buffer, including the text under excised nodes.
$text	N, X	Fetch the text covered by the node. Cleanup whitespace (removing leading and trailing, convert separators to single space). Uses the original text buffer, rather than the subtree under the node, in order to gather text. See also $treetext and $treeraw.
$treeraw	N, X	Fetch the text covered by the node from the parse tree. Text is not 'cleaned up'. Analogous to $treetext. See also $treetext and $raw.
$raw	N, X	Fetch the 'cleaned up' text covered by the node from the text buffer. See also $treetext, $treeraw and $text.
$xmltext	N, X	Same as $raw, but converts characters that are special to HTML and XML. E.g., "<" is converted to "<".
$length	N, X	Get the length of node's text.
$ostart	N, X	Start offset of the referenced node in the input text.
$oend	N, X	End offset of the referenced node in the input text.
$start	N, X	Evaluates to 1 if the referenced node has no left sibling in the parse tree, else 0.
$end	N, X	Evaluates to 1 if the referenced node has no right sibling in the parse tree, else 0.
$apppath	G	Get base directory path for current application. E.g., "D:\dev\apps\Resume".
$input	G	Get full input file. E.g., "D:\apps\Resume\input\Dev1\rez.txt"
$inputpath	G	Get input file path. E.g., "D:\apps\Resume\input\Dev1"
$inputname	G	Get input filename. E.g., "rez.txt"
$inputhead	G	Get input file head. E.g., "rez"
$inputtail	G	Get input file tail (i.e., extension). E.g., "txt"
$allcaps $uppercase	N	Returns 1 if the token underlying the node is all uppercase. If multiple words (even if all are all-caps), returns 0. WARN: Not identical to the uppercase() pre action.
$lowercase	N	Returns 1 if the token underlying the node is all lowercase If multiple words (even if all are all-caps), returns 0. WARN: Not identical to the lowercase() pre action.
$cap	N	Returns 1 if the token underlying the node is a capitalized word. WARN: Not identical to the cap() pre action.
$mixcap	N	Returns 1 if the token underlying the node is a mixed-capitalized word. Examples: "MIchigan", "abcD".
$exists	N	Returns 1 if the node exists, 0 otherwise.
$unknown	N	Returns 1 if the token underlying the node is an unknown word. WARN: Not identical to the unknown() pre action. Requires a lookup() code action prior to any use of this special variable.

NOTE:
[1] Special variable names must be quoted. E.g., N("$text", 2).
[2] You should not assign to or create your own "dollar" variables. NLP++ reserves all variable names that start with a dollar sign.
[3] The $uppercase, $lowercase, $cap, and $unknown differ from the analogous pre actions in that they all traverse down to the "leafiest" node they can find, in order to perform their check. The corresponding pre actions only chain down if the [s] flag is present in the associated phrase element(s). Also, the $ variables are not stopped by a [base] attribute in a node.

Parse Tree Nodes

NLP++ can also reference the nodes of a parse tree directly. For example, N(2) refers to the (first) node that matched the 2nd element of a rule's right-hand-side phrase. X(3) refers to the 3rd node in the current context path as specified by @PATH. These constructs are used in functions that deal with the parse tree (see the Parse Tree Functions section below). (Note that knowledge base nodes are unrelated to parse tree nodes. The former are called NODE and the latter PNODE, as in the Types section below.)

Types

The types of NLP++ variables available are shown in the following table.

NAME	ABBREVIATION	CHARACTERISTICS
string	STR	Alpha-numerics: e.g. G("foobar") = "w3c";
integer	INT	32-bit integers: e.g. G("foorbar") = 32776;
boolean	BOOL	1 or 0 (Boolean implemented as a numeric 1 for true and the numeric 0 for false.)
ostream	OSTREAM	C++ output stream. fileout(filename) creates a variable with ostream value and filename << expr; prints by referring to the ostream value of G(filename). See also the ofile NLP++ function.
concept	CON	Concepts are the main object in the kb and appear in the kb hierarchy. They have string names, and optionally have one or more attributes. e.g. G("myConcept") = makeconcept(G("root"),"My Name");
attribute	ATTR	Attributes are associated with concepts and kb nodes. They have string names, and optionally have one or more values. e.g. G("myAttribute") = addattr(G("myConcept"),"is");
value	VAL	Values are the values of kb attributes. They can be strings, integers or concepts: e.g. addconval(G("myConcept"), "is", "John Smith");
phrase	PHR	A phrase is a sequence of nodes owned by a single concept. E.g., create the first node of a phrase: G("USPhrase") = makephrase(G("US"),"United"); Add the last node of the phrase: addcnode(G("US"), "States");
node	NODE	Node in the KB. Differs from a concept in that it is not in the KB hierarchy, but rather in a concept's phrase. A node can serve as a "proxy" for a concept.
array	ARRAY	The values of an array, or multi-valued variable, can be any single value (i.e., any type other than a nested ARRAY).
pnode	PNODE	Node in the parse tree. Unrelated to a KB node.

Arrays

Arrays can consist of combinations of simple types. Nested array types are not implemented. An example follows.

# Create an array and step through it.
G("counter") = 0;
G("array_size") = 10;
# Arrays can optionally be "declared" by padding them with 0.
G("int_array")[G("array_size") - 1] = 0;

while ( G("counter") < G("array_size") ) {
G("int_array")[G("counter")] = G("counter");
"output.txt" << G("int_array")[G("counter")] << "\n";
G("counter") ++;
}
# print it out all at once
"output.txt" << G("int_array") << "\n";.
# copy it all at once:
G("another_array") = G("int_array");

Operators

Most NLP++ operators are modeled on the C programming language. The main addition is the confidence operator, %%. This enables convenient accumulation of evidence in terms of percent confidence. Following is an operator precedence table, with highest precedence at the top.

OPERATOR	DESCRIPTION	ASSOCIATIVITY
++ --	Post increment, decrement	Left to right
++ --	Pre increment, decrement	Right to left
!	Logical NOT (unary)	Right to left
+ -	Unary plus, minus	Right to left
*% / %%**	Remainder, multiplication, division, confidence	Left to right
+ -	Addition, subtraction	Left to right
< > <= == != >=	Relational operators	Left to right
&& \|\|	Logical AND, OR	Left to right
=	Assignment	Right to left (multiple assignment works)
*= /= += -= %%=	Shorthand assignment (unimplemented)	Right to left
<<	Output operator	Left to right

Output Operator

The << output operator is modeled on the C++ operator. Only the following form is supported, currently:

filename << expression << expression .... ;

Example:

"output.txt" << "Hello, NLP++!" << "\n";

NOTES: The file named by filename should have been created by a prior fileout(filename) code action. If that hasn't been done, then the output operator will create the global variable G(filename) with the output stream as its value. The default output mode for the file is append, which can't be modified by the user at present. Also, the default output directory is the log directory for the current input file -- also not under user control at present.

Confidence Operator

Say that something in the parse tree gives you 80% confidence that you've found a name. You could write this into a global variable as

G("name confidence") = 80;

Now you want to account for a new piece of evidence. Say the standalone confidence of that evidence is 70%. You could combine this as

G("name confidence") = G("name confidence") %% 70;

which will yield a new percent confidence between 80% and 100%.

The %% operator guarantees that "adding" or "subtracting" confidence will always stay within the range of -100 to +100, providing an intuitive way to gather evidence and compare confidences using a single scale of measurement. (Actually, %% bottoms out at 0%. A planned change to the operator will enable it to bottom out at -100%, which can be thought of as 100% confidence against something.)

Expressions

NLP++ expressions are modeled on the C/C++ programming languages and can be nested arbitrarily and grouped using parentheses ( ).

One useful enhancement: addition with string arguments or string variables equates to string catenation.

Functions and Actions

NLP++ supports function calls. Actions, e.g., single, are specialized functions that don't take generalized NLP++ expressions as arguments (see Pass Files in the Developers section). Currently, new NLP++ functions must be defined with C++ code, within an application's user project. User-defined functions must be preceded by the 'user' scope qualifier, e.g., user::myfunction. (Interpreted NLP++ will find the correct function and issue a warning, but user functions without the scope qualifier will be compiled incorrectly.)

Statements and Blocks

This too is modeled after the C programming language. Curly braces { and } delimit blocks of statements. Statements should end with a semicolon (;). A statement can be an arbitrary expression, a block, or fully nested if-else statements.

While Loop

The only loop construct current supported is the while loop. Here's an example that prints out 0 thru 9:

G("counter") = 0;
G("loop_limit") = 10;
while( G("counter") < G("loop_limit")) {
"output.txt" << "Loop counter = " << G("counter") << "\n";
G("counter")++;
}

Parse Tree Functions

FUNCTION NAME	RETURNS	DESCRIPTION
arraylength(VAR)	INT	Return number of values in given variable or expr. (Like a call-by-reference)
lasteltnode(elt_num)	PNODE	Get the last node that matched the elt_numth element of a rule.
phrasetext()	STR	Get the text that matched the right hand side phrase of a rule. Analogous to $text.
phraseraw()	STR	Get the raw text that matched the right hand side phrase of a rule. Analogous to $raw
pndeletechilds(pnode)	None	Delete children of given pnode.
pnname(pnode)	STR	Return pnode's name.
pnnext(pnode)	PNODE	Return pnode's right sibling, if any.
pnprev(pnode)	PNODE	Return pnode's left sibling, if any.
pnrename(pnode, str)	STR	Rename given pnode. Returns the interned name.
pnroot()	PNODE	Return root of parse tree
pnsingletdown(pnode)	PNODE	Return pnode's child, obeying rules about going down a "singlet chain." I.e., get child only if no branching and if not going past node with BASE flag set.
pnup(pnode)	PNODE	Return pnode's parent, if any. Only leftmost pnode in a list has a parent.
pnvar(pnode, var_str)	(var values)	Get pnode's variable value(s).
varinlist(var_str, elt_num)	(var values)	Get var_str variable's value(s) in any pnode that matched elt_numth element

String Functions

FUNCTION NAME	RETURNS	DESCRIPTION
strchar(str, num)	STR	Index to numth char of string
strchr(str, ch_str)	STR	Find first occurrence of char in string. Return string headed by char.
strclean(str)	STR	Remove leading, trailing, and repeat whitespace separators.
strcontains(str1,str2)	BOOL	Is str1 contained in str2.
strcontainsnocase(str1,str2)	BOOL	Is str1 contained in str2, ignoring letter case.
strendswith(str, suffix_str)	BOOL	Does given str end with given suffix_str
strequal(str1,str2)	BOOL	Is str1 identical to str2.
strisalpha(str)	BOOL	Is str all alphabetic chars.
strisdigit(str)	BOOL	Is str all numeric chars.
strnotequal(str1,str2)	BOOL	Does str1 differ from str2.
strequalnocase(str1,str2)	BOOL	Is str1 identical to str2, ignoring letter case.
strnotequalnocase(str1,str2)	BOOL	Does str1 differ from str2, ignoring letter case.
strlessthan(str1,str2)	BOOL	Is str1 lexically before str2.
strgreaterthan(str1,str2)	BOOL	Is str1 lexically after str2.
strlength(str)	INT	Return length of string.
strpiece(str, start_num, end_num)	STR	Return substring from start_num to end_num indexes. Zero-based
strrchr(str, ch_str)	STR	Find last occurrence of char in string. Return string headed by char
strsubst(str, old_str, new_str)	STR	In str, substitute instances of old_str with new_str.
strtolower(str)	STR	Return lowercase form of string.
strtotitle(str)	STR	Capitalize string as in a title or header.
strtoupper(str)	STR	Return uppercase form of string
strtrim(str)	STR	Remove leading and trailing whitespace from str.
unpackdirs(dir_str)	STR ARRAY	Unpack directory names from a full directory path string.
strwrap(str, num)	STR	Fold string successively at every numth character.
num(str)	NUM	Convert string to number, if possible. (Also accepts num arg).
str(num)	STR	Convert numeric value to string. (Also accepts string arg).

Miscellaneous Functions

FUNCTION NAME	RETURNS	DESCRIPTION
fail()	None	In @CHECK region, abort the rule that has just matched. Noop elsewhere.
succeed()	None	In @CHECK and @CODE region, succeed without performing further code actions in that region.
batchstart()	BOOL	True if current file is first in a batch run.
debug()	BOOL	No op. Convenient breakpoint for debugging from C++ compiler.
hitconf(hits_num, total_num, fudge_num)	0 to 100	Percent confidence for hits. Fudge factor is typically 3 to 20.
LJ(num, fieldsize_num)	STR	Left-justify num in given field size.
ofile(path_str [, mode_str [, mode_str [, mode_str]]] )	OSTREAM	Open an output file for the given path; analogous to C++ open file function. Zero to three mode strings selected from: 1) "app" = append; 2) "ate" = special C++ append; 3) "binary" = open in binary mode (text is default). WARNINGS: Default mode is OVERWRITE, not APPEND. Unlike fileout(), does not create a global variable for the file stream. [Upgraded doc to conform to standard C++ i/o libraries.]
percentstr(numerator_num, denominator_num)	STR (" 0" to "100")	Format a percentage (right-justified in field of 3 chars).
rightjustifynum(num, fieldsize_num)	STR	Format a percentage (right-justified in field of 3 chars).
spellcandidates(word_str)	STR	Return list of space separated candidates for spell-correcting given word.
spellcorrect(word_str)	STR	If known word, return as-is. If not, return best spelling guess.
spellword(word_str)	BOOL	See if given word is a known English word.
strspellcandidate(word_str, list_str)	STR	Choose best spell-correct candidate for a word, given a list of space-separated candidates (as returned by spellcandidates, e.g.)
strspellcompare(str1, str2)	INT	Return spelling "distance" between two given words. The smaller the number, the fewer the corrections to get from one to the other.
system(str)	BOOL	Execute str as operating system command

Note: It is necessary to escape the directory delimiter backslash ('\') with another ('\'). For example, suppose you wanted to echo characters to a file and then copy that file to some other directory. Here's some example code:

System Function Example

@NODES _ROOT
@POST
# echo.txt is put in your default console location, which for
# me happens to be D:\WINNT\Profiles\Administrator\Desktop
system("echo this is a test > echo.txt");
# echo1.txt ends up in D:\WINNT\Profiles\Administrator\Desktop
system("copy echo.txt echo1.txt");
# echo2.txt ends up on D:
system("copy echo1.txt D:\\echo2.txt");
# echo3.txt ends up on D:\apps
system("copy D:\\echo2.txt D:\\apps\\echo3.txt");
"output.txt" << "copy.\n" ;
noop();
@@POST
@RULES
_xNIL <- _xALPHA @@