This page describes how to build custom recognition resources to fine-tune the recognition.

How to build resources

This is a two-step process: You should first prepare a resource file. This file is a UTF-8-encoded file containing dedicated resources information.

For Text recognition, possible resources are lexicon and subset knowledge. Lexicon files should have a .txt or a .lex extension, whereas subset knowledge files should have a .txt or a .sk one.

For Math recognition with a math bundle, you can set up your math grammar defining the symbols and rules. They shall be saved in a .txt or .def file.

For Math recognition with a math2 bundle, you can set up your math subset knowledge defining the symbols and rules, and compile it on device.

Once you have written your resource file, you must compile it to generate the corresponding binary .res file.

How to compile resources

MyScript Developer Portal comes with an online tool that lets you compile your own resources files.

Text

Lexicon

The lexicon is a list of words or expressions that is being used by the engine to recognize a specific set of terms. You can create your own lexicon to improve the recognition. It may contain terms that are unlikely to appear in a classical dictionary but that you will be led to write many times (proper nouns like the name of your company or your employees, your password, a hashtag, etc.).

For example, a lexicon can be the following set:

Johnson
Meyer
Lopez
Gibbons
Cooper
Martin
Bailey

If needed, you can also download another Lexicon example.

You can attach several lexicons to a configuration to complement or replace the MyScript-provided LK.

Subset knowledge

The Subset Knowledge resource (or SK) is a white list that works as a filter to constrain the recognition to a specific set of characters. You can create your own SK to improve the recognition and help the engine reduce the “margin of error”. The expected set can be limited to digits or to letters in a given alphabet or language.

For example: In a form field where you expect a phone number, constraining the recognition with a SK could avoid the engine to mistake a “0” for a “O” or a “1” for a “l”.

See below examples of SK resources:

0123456789

If needed, you can download another SK example.

Math

Grammar resource

The grammar resource indicates the way to parse handwritten mathematical expressions. It specifies:

a limited set of terminal symbols,
Terminal symbols are the elementary symbols. They are basic symbols such as a, b, c, …, 0, 1, …, +, -, ±, etc. that cannot be broken down into “smaller” recognized units. See below the list of supported symbols.
a limited set of non-terminal symbols,
Non-terminal symbols describe groups of terminal symbols organized according to rules.
a limited set of rules,
Rules describe the way to parse digital ink. For instance, a fraction is a rule that contains a numerator, a fraction bar and a denominator. The recognizer expects to find these three elements to fit the fraction rule. See below the list of supported rules.
a start symbol.
This defines the way your mathematical expression should be read depending on the above elements.

For example, a grammar resource can be:

symbol = 0 1 2 3 4 5 6 7 8 9 + - / ÷ = . , % | ( ) : * x
leftpar = (
rightpar = )
currency_symbol = $ R € ₹ £
character ::= identity(symbol)
            | identity(currency_symbol)
fractionless ::= identity(character)
               | fence (fractionless, leftpar, rightpar)
               | hpair(fractionless, fractionless)
fractionable ::= identity(character)
               | fence (fractionable, leftpar, rightpar)
               | hpair(fractionable, fractionable)
               | fraction(fractionless, fractionless)
expression ::= identity(character)
             | fence (expression, leftpar, rightpar)
             | hpair(expression, expression)
             | fraction(fractionable, fractionable)
start(expression)

If needed, you can download another Grammar example.

You can only attach one grammar per configuration.

Grammar syntax

You can build your own MyScript-Math-compliant grammar resource. To do so, you must define the grammar in conformity with the syntax set out below and then compile it.

You can add comments in a grammar resource as follows:

// This is the first way to start a whole comment line.
#  This is the second way to start a whole comment line.
"  This is the third way to start a whole comment line.
/* This is a block comment. */

Inline comments are not supported.

Your grammar resource will contain terminal symbol definitions such as the following:

my_terminal_name = 0 1 2 3 4 5 6 7 8 9

The terminal symbol name is defined as: [-a-zA-Z_][-a-zA-Z_0-9]
It is a character string which starts with a character from [-a-zA-Z_] and whose other characters are any of [-a-zA-Z_0-9]. In the above example, the terminal symbol name is my_terminal_name.
The list of symbols referred to by the terminal name are defined as: ( ( !EOL . )+ | EOL space+ )+
The definition of the symbols referred to by the terminal name does not start by a EOL (end of line). They can be any existing character separated by a space or EOL and a space. In other words, the definition of the symbols is allowed to span multiple lines, providing that the “continuation line” starts with a space. This means you can format long terminal symbol definitions in a more visually organized (“pretty-formatted”) manner. In the above example, the list of symbols referred to by my_terminal_name is 0 1 2 3 4 5 6 7 8 9.
Space is defined as ‘ ’ or ‘\t’
EOL is defined as ‘\r\n’ or ‘\n’ or ‘\r’

You will also need to list non-terminal symbol definitions. Here is an example:

my_non_terminal_name ::= fraction(my_terminal_name,my_terminal_name)

The non-terminal symbol name is defined as : [-a-zA-Z_][-a-zA-Z_0-9]
It is a character string which starts with a character from [-a-zA-Z_] and whose other characters are any of [-a-zA-Z_0-9]. In the above example, the non-terminal symbol name is my_non_terminal_name.

The non-terminal symbol is defined as: non_terminal_name ::= rule (| rule)?

In the above example, the rule is fraction. The example specifies that the numerator is a my_terminal_name symbol and that the denominator is also a my_terminal_name symbol.

Putting this example aside, note that MyScript Math also offers the possibility to recognize “partial” fraction, i.e. with one denominator only or with one numerator only. This feature is available by default, so that you do not need to add a non-terminal entry to recognize “partial” fractions.

Rule continuations allow you to “pretty format” rule definitions by avoiding a repetition of target non-terminal symbol names.

Finally, your grammar must include the start symbol definition:

start(my_non_terminal_name)

Finally, you must define the start symbol. In the above example, we specify that the general mathematical expression form that will be recognized is my_non_terminal_name. In other words, we expect the input digital ink to represent a fraction of digits.

Here is an example of a custom grammar:

Supported symbols and rules

A non-exhaustive list of supported math symbols and rules can be found here.

It is also possible to create a custom grammar resource, by constraining the recognition for particular use cases (integral calculus, vector calculus, finite element calculus, etc.).

The following table provides for each supported math rule its denomination in the grammar as well as the parameters it supports:

Rule	Visual structure	Syntax	Explanation
Identity	N/A	`identity(source)`	Reuse a previously defined symbol in a rule clause
Horizontal pair		`hpair(left, right)`	Ordered juxtaposition of the `left` and `right` expressions
Fence		`fence(exp, left, right)`	`exp` is placed between the `left` and `right` symbols
Left fence		`leftfence(exp, symbol)`	`symbol` is positioned at the left of `exp`. This allows for instance defining systems
Square root		`sqrt(exp)`	`exp` is placed under the square root
Fraction		`fraction(numerator, denominator)`	`numerator` is divided by `denominator`
Subscript		`subscript(exp, index)`	`index` is placed as subscript at the right of `exp`
Superscript		`superscript(exp, exponent)`	`exponent` is placed as superscript at the right of `exp`
Subsuperscript		`subsuperscript(exp, index, exponent)`	`index` and `exponents` are respectively placed as subscript and superscript at the right of `exp`
Presuperscript		`presuperscript(exp, exponent)`	`exponent` is placed as superscript at the left of `exp`
Overscript		`overscript(exp, top)`	`top` is placed above `exp`
Underscript		`underscript(exp, bottom)`	`bottom` is placed as subscript at the left of `exp`
Underoverscript		`underoverscript(exp, bottom, top)`	`top` and `bottom` are respectively placed as superscript and subscript at the left of `exp`
Vertical pair		`vpair(top, bottom)`	`top` and `bottom` are respectively placed one on top of the other
Vertical list		`vlist(exp)`	`exp` represent the expressions that can be placed on several consecutive lines
Table		`table(exp)`	`exp` represent the expressions that can be placed inside the table cell. This allows for instance defining matrices
Partial fraction (numerator)		`partialfractionnumerator(numerator)`	Only the numerator of the fraction is defined
Partial fraction (denominator)		`partialfractiondenominator(denominator)`	Only the denominator of the fraction is defined
Slanted fraction		`slantedfraction(numerator, denominator)`	`numerator` is divided by `denominator`

Math Subset Knowledge

The Subset Knowledge resource (or SK) is a white or black list that works as a filter to constrain the recognition to a specific set of symbols and rules for Math Recognizer objects or math editor using math2 bundle.

You can define your SK and then use it, with either "Math Enabled Subset" option to whitelist it, or "Math Disabled Subset" to blacklist it. The example below illustrated a SK resource definition:

0
1
2
3
4
5
6
7
8
9
+
-
/
÷
.
,
=
(
)
%
SqrtRule
FracRule

A non-exhaustive list of supported math symbols and rules can be found here.

The following section describes how to build a math SK for native usage. For the Web usage, refer to the REST recognizer section.

You can also get the exhaustive list of math and and rules symbols that is supported by the SK via the GetSupportedSymbols() method of the RecognitionAssetsBuilder object.