Grammalecte  Check-in [f39d633667]

# Writing rules for Grammalecte

Note: This documentation is a __draft__. Information may be obsolete or incomplete.


## FILES REQUIRED

The rules file for your language must be named `rules.grx` in the folder `gc_lang/<lang>/`.
The settings file must be named `config.ini`.

These files are simple UTF-8 text files.


## 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.

You may alter how sentences are split by removing punctuation marks during the first pass.

* a list of actions

A token rule is defined by:

* rule name
* [optional] priority number
* one or several lists of tokens
* a list of actions (an 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|graph_code

A graph ends when another graph is defined or when is found 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 several kinds of actions:

* Error warning, with a message, and optionally suggestions, and optionally an URL
* Text transformation, modifying internally the checked text
* Disambiguation action
* [second pass only] Tagging token
* [second pass only] Immunity rules

On the first pass, you can only write regex rules.
On the second pass, you can write regex rules and token rules. All token rules must be written within a graph.


## REGEX RULE SYNTAX

    __LCR/option(rulename)!priority__
        pattern
            <<- condition ->> error_suggestions         && message_error|URL
            <<- condition ~>> text_rewriting
            <<- condition =>> commands_for_disambiguation
            ...

>   `i`     case insensitive

>   `s`     case sensitive

>   `u`     uppercase allowed for lowercase characters

> >          i.e.:  "Word"  becomes  "W[oO][rR][dD]"

Examples: `[i]`, `<s]`, `[u>`, `<s>`

User option activating/disactivating is possible with an option name placed
just after the LCR flags, i.e.:

    __[i]/option1(rulename1)__
    __[u]/option2(rulename2)__
    __[s>/option3(rulename3)__
    __<u>(rulename4)__
    __<i>(rulename5)__

Each rule name must be unique.

Example. Recognize and suggest missing hyphen and rewrite internally the text
with the hyphen:

    __[s](rulename)__
        foo bar
            <<- ->> foo-bar     && Missing hyphen.
            <<- ~>> foo-bar


### Simple-line or multi-line rules

Rules can be break to multiple lines by leading spaces.
You should use 4 spaces.

Examples:

    __<s>(rulename)__ pattern <<- condition ->> replacement  && message
    
    __<s>(rulename)__
        pattern
            <<- condition ->> replacement
            && message
            <<- condition ->> suggestion  && message
            <<- condition ~>> text_rewriting
            <<- =>> disambiguation


### Whitespaces at the border of patterns or suggestions

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

    __<s>(rulename)__  "  +" <<- ->> " "      && Remove extra space(s).

Characters `"` protect spaces in the pattern and in the replacement text.


### Pattern groups and back references

It is usually useful to retrieve parts of the matched pattern. We simply use
parenthesis in pattern to get groups with back references.

Example. Suggest a word with correct quotation marks:

    \"(\w+)\" <<- ->> “\1”      && Correct quotation marks.

Example. Suggest the missing space after the signs `!`, `?` or `.`:

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

Example. Back reference in messages.

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


### Group positioning codes for JavaScript:

There is no way in JavaScript to know where a captured group starts and ends. To avoid misplacement, regex rules may specify group positioning codes which indicate to the grammar checker where is the position of the captured groups.

A group positioning code always begins by `@@`. If there is several codes, they are separated by a comma `,`.

    **                  something somewhere after previous group

Examples:

    ([A-ZÉÈÂÎ][\w-]+) [A-ZÉÈÂ]([.]) ([A-ZÉÈÂ][\w-]+)  @@0,*,$
    " ([?!;])"  @@1


### 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


## TOKEN RULES

Token rules must be defined within a graph.

### Token rules syntax

    __rulename!priority__
        list_of_tokens

Tokens can be defined in several ways:

* Value (the text of the token). Examples: `word`, `<start>`, `<end>`, `,`.
* Lemma: `>lemma`.
* Regex: `~pattern`, `~pattern¬antipattern`.
* Regex on morphologies: `@pattern`, `@pattern¬antipattern`.
* Tags: `/tag`.
* Metatags: `*METATAG`. Examples: `*WORD`, `*NUM`, `*SIGN`, etc.
* Jump over token: `<>`

Selection of tokens: `[value1|value2|>lemma|~pattern|@pattern|*META|/tag|…]`

Conditional token: `?token¿`

Conditional selection of token: `?[token1|token2|…]¿`

### Token references

Positive references are defined by a positive integer (> 0). Examples: `\1`, `\2`, `\3`, etc.
If there is at least one token set between parenthesis, these numbers refer to tokens between parenthesis, ignoring all others.
If there is no token between parenthesis, these numbers refer to tokens found in order defined by the rule triggered.

Negative references are defined by a negative integer (< 0). Examples: `\-1`, `\-2`, `\-3`, etc.
These numbers refer to the tokens beginning by the last one found by the rule triggered.

Examples:

    tokens:             alpha       beta        gamma       delta       epsilon
    positive refs:      1           2           3           4           5
    negative refs:      -5          -4          -3          -2          -1
    
    tokens:             alpha       (beta)      gamma       (delta)     epsilon
    positive refs:                  1                       2
    negative refs:      -5          -4          -3          -2          -1
    
    tokens:             alpha       (beta)      ?gamma¿     (delta)     epsilon
    positive refs:                  1                       2
    negative refs:      (-5/-4)     (-4/-3)     (-3/none)   -2          -1


## CONDITIONS

Conditions are Python expressions, they must return a value, which will be
evaluated as boolean. You can use the usual Python syntax and libraries.

With regex rules, you can call pattern subgroups via `\1`, `\2`… `\0` is the full pattern.

Example:

    these (\w+)
        <<- \1 == "man" -1>> men        && Man is a singular noun.

You can also apply functions to subgroups like: `\1.startswith("a")` or `\3.islower()` or `re.search("pattern", \2)`.

With token rules, you can also call each token with their reference, like `\1`, `\2`... or `\-1`, `\-2`...

Example:

    foo [really|often|sometimes] bar
        <<- ->> \1 \-1                  && We say “foo bar”.


### Functions for regex rules

`word(n)`

>   Catches the nth next word after the pattern (separated only by white spaces).
>   Returns None if no word caught

>   Returns True if <tag> is found any token after the nth tag.

### Functions for regex and token rules

`__also__`

>   Returns True if the previous condition returned True.
>   Example: `<<- __also__ ->>`
>   Example: `<<- __also__ and condition2 ->>`

`__else__`

>   Returns False if the previous condition returned False.
>   Example: `<<- __else__ ->>`
>   Example: `<<- __else__ and condition2 ->>`

`option(option_name)`

>   Returns True if <option_name> is activated else False

Note: the analysis is done on the preprocessed text.

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

`sContext`

>   The name of the application running (Python, Writer…)


## ACTIONS

There are 5 kinds of actions:

1. Suggestions. The grammar checker suggests corrections.
2. Text processor. A internal process to modify the text internally. This is used to simplify grammar checking.
   * text rewriting
   * text deletion
   * token rewriting
   * token merging
   * token deletion
3. Disambiguation. Select, exclude or define morphologies of tokens.
4. Tagging. Add information on token.
5. Immunity. Prevent suggestions to be triggered.


### Positioning

Positioning is valid for suggestions, text processing, tagging and immunity.

By default, rules apply on the full text triggered. You can shorten the
effect of rules by specifying a back reference group of the pattern or token references.

Instead of writing `->>`, write `-n>>`  n being the number of a back reference
group. Actually,  `->>`  is similar to  `-0>>`.

Example:

    (ying) and yang <<- -1>> yin   && Did you mean:
    
    __[s]__ (Mr.) [A-Z]\w+ <<- ~1>> Mr


**Comparison**

Rule A:

    ying and yang       <<- ->>     yin and yang        && Did you mean:

Rule B:

    (ying) and yang     <<- -1>>    yin                 && Did you mean:

With the rule A, the full pattern is underlined:

    ying and yang
    ^^^^^^^^^^^^^

With the rule B, only the first group is underlined:

    ying and yang
    ^^^^


### Errors and suggestions

The command to suggest something is: `->>`.

#### Multiple 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
    <<- ~>> =\1.upper()


### Text rewriting

**WARNING**: The replacing text must be shorter than the replaced text or have the same length. Breaking this rule will misplace following error reports.
You have to ensure yourself the rules comply with this constraint, the text processor won’t do it for you.

The command for text rewriting is: `~>>`.

You can use positioning with text rewriting actions.

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

You can also call Python expressions.

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


The text processor is useful to simplify texts and write simpler checking
rules.

For example, sentences with the same grammar mistake:

    These “cats” are blacks.

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 the grammar checker analyses a token with `morph()`, before requesting the POS tags to the dictionary, it checks if there is a stored marker for the position of the token. If a marker is found, it uses the stored data and don’t make request to the dictionary.

The command for disambiguation is: `=>>`. No positioning allowed.

**Only for token rules**. A immunity rule set a flag on token(s) who are not supposed to be considered as an error. If any other rules find an error, it will be ignored. If an error has already been found, it will be removed.

Example: `!2>>` means no error can be set on the second token.
Example: `!>>` means all tokens will be considered as correct.

The immunity rules are useful to create simple antipattern that will simplify writing of other rules.


## OTHER COMMANDS

### Comments

Lines beginning with `#` are comments.

### End of parsing

Usage: `{name}` will be replaced by its definition

Example:

    DEF: word_3_letters     \w\w\w+
    DEF: uppercase_token    ~^[A-Z]+$
    DEF: month_token        [January|February|March|April|May|June|July|August|September|October|November|december]
    
    ({word_3_letters}) (\w+) <<- condition ->> suggestion     && message|URL
    
    {uppercase_token} {month_token}
        <<- condition ->> message                             && message|URL