Grammalecte  Diff

WRITING RULES FOR GRAMMALECTE

Note: This documentation is a draft. Information may be obsolete.

# Principles #

Grammalecte is a bi-passes grammar checker engine. On the first pass, the
engine checks the text paragraph by paragraph. On the second pass, the engine
check the text sentence by sentence.

The command to switch to the second pass is `[++]`.

In each pass, you can write as many rules as you need.

There is two kinds of rules:

* regex rules (triggered by a regular expression)
* token rules (triggered by a succession of tokens)

A regex rule is defined by:

* [optional] flags “LCR” for the regex word boundaries and case sensitiveness
* a regex pattern trigger
* a list of actions
* [optional] option name (the rule is active only if the option defined by user or config is active)
* [optional] rule name (named rules can be disabled by user or by config)

A token rules is defined by:

* rule name
* one or several lists of tokens (triggers)
* a list of actions (the action is active only if the option defined by user or config is active)

Token rules must be defined within a graph.

Each graph is defined within the second pass with the command:

        @@@@GRAPH: graph_name

A graph ends when another graph is defined or when is defined the command:

        @@@@END_GRAPH

There is no limit to the number of actions and the type of actions a rule can
launch. Each action has its own condition to be triggered.

There are three kind of actions:

* Error warning, with a message, and optionally suggestions, and optionally an URL
* Text transformation, modifying internally the checked text
* Disambiguation action, setting tags on a position


The rules file for your language must be named “rules.grx”.
The settings file must be named “config.ini”.

All these files are simple utf-8 text file.
UTF-8 is mandatory.


# Comments #

Lines beginning with `#` are comments.


# End of file #

With the command:

        #END

at the beginning of a line, the parser won’t go further.
Whatever is written after will be considered as comments.


# Regex rule syntax #

        __LCR/option(rulename)__
            pattern
            <<- condition ->> error_suggestions  # message_error|http://awebsite.net...
            <<- condition ~>> text_rewriting
            <<- condition =>> commands_for_disambiguation
            ...

Patterns are written with the Python syntax for regular expressions:
http://docs.python.org/library/re.html

Examples:

        __<s>__ pattern
            <<- condition ->> replacement
            # message
            <<- condition ->> suggestion # message
            <<- condition ~>> text_rewriting

            <<- =>> disambiguation

        __<s>__ pattern <<- condition ->> replacement # message

















## Whitespaces at the border of patterns or suggestions ##

Example: Recognize double or more spaces and suggests a single space:

        __<s>__  "  +" <<- ->> " "      # Extra space(s).

        __<i]__ \b([?!.])([A-Z]+) <<- ->> \1 \2     # Missing space?

Example. Back reference in messages.

        (fooo) bar <<- ->> foo      # “\1” should be:


## Pattern matching ##

Repeating pattern matching of a single rule continues after the previous matching, so
instead of general multiword patterns, like

(\w+) (\w+) <<- some_check(\1, \2) ->> \1, \2 # foo

use

(\w+) <<- some_check(\1, word(1)) ->> \1, # foo


## Name definitions ##

Grammalecte supports name definitions to simplify the description of the
complex rules.

Example:

        Mr(. [A-Z]\w+) <<- ~1>> *

You can also call Python expressions.

        __[s]__ Mr. ([a-z]\w+) <<- ~1>> =\1.upper()


# Text preprocessing and multi-passes checking #

On each pass, Lightproof uses rules written in the text preprocessor to modify
internally the text before checking the text.

The text preprocessor is useful to simplify texts and write simplier checking
rules.

For example, sentences with the same grammar mistake:

        These “cats” are blacks.
        These cats are “blacks”.
        These cats are absolutely blacks.
        These stupid “cats” are all blacks.
        These unknown cats are as per usual blacks.

Instead of writting complex rules or several rules to find mistakes for all possible
cases, you can use the text preprocessor to simplify the text.

To remove the chars “”, write:

        [“”] ->> *

The * means: replace text by whitespaces.

Similarly to grammar rules, you can add conditions:

        \w+ly <<- morph(\0, "adverb") ->> *

You can also remove a group reference:

        these (\w+) (\w+) <<- morph(\1, "adjective") and morph(\2, "noun") -1>> *
        (am|are|is|were|was) (all) <<- -2>> *

With these rules, you get the following sentences:

        These  cats  are blacks.
        These cats are  blacks .
        These cats are            blacks.
        These         cats  are     blacks.
        These         cats are              blacks.

These grammar mistakes can be detected with one simple rule:

        these +(\w+) +are +(\w+s)
            <<- morph(\1, "noun") and morph(\2, "plural")
            -2>> _              # Adjectives are invariable.

Instead of replacing text with whitespaces, you can replace text with @.

        https?://\S+ ->> @

This is useful if at first pass you write rules to check successive whitespaces.
@ are automatically removed at the second pass.

You can also replace any text as you wish.

        Mister <<- ->> Mr
        (Mrs?)[.] <<- ->> \1


# Disambiguation #

When Grammalecte analyses a word with morph or morphex, before requesting the
POS tags to the dictionary, it checks if there is a stored marker for the
position where the word is. If there is a marker, Grammalecte uses the stored
data and don’t make request to the dictionary.

>   checks if the text before the pattern matches the regex.

`textarea(regex[, neg_regex])`

>    checks if the full text of the checked area (paragraph or sentence) matches the regex.

`morph(n, regex[, neg_regex][, no_word=False])`

>   checks if all tags of the word in group n match the regex.
>   if neg_regex = "*", returns True only if all morphologies match the regex.
>   if there is no word at position n, returns the value of no_word.

`analyse(n, regex[, neg_regex][, no_word=False])`

>   checks if all tags of the word in group n match the regex.
>   if neg_regex = "*", returns True only if all morphologies match the regex.
>   if there is no word at position n, returns the value of no_word.


`option(option_name)`

>   returns True if option_name is activated else False

Note: the analysis is done on the preprocessed text.


# Default variables #

`sCountry`

>   It contains the current country locale of the checked paragraph.

        colour <<- sCountry == "US" ->> color       # Use American English spelling.



# Expressions in the suggestions #

Suggestions started by an equal sign are Python string expressions
extended with possible back references and named definitions:

Example:

        <<- ->> = '"' + \1.upper() + '"'     # With uppercase letters and quotation marks















# Token rules


Token rules must be defined within a graph.








## Tokens






















Tokens can be defined in several ways:





* Value (meaning the text of the token). Examples: `word`, `<start>`, `<end>`, `,`.

* Lemma: `>lemma`

* Rege: `~pattern`


* Regex on morphologies: `@pattern`, `@pattern¬antipattern`.

* Metatags: *NAME. Examples: `*WORD`, `*SIGN`, etc.