NAME

Mail::SpamAssassin::Plugin::TextCat - TextCat language guesser

SYNOPSIS

loadplugin     Mail::SpamAssassin::Plugin::TextCat

DESCRIPTION

This plugin will try to guess the language used in the message body text.

You can use the "ok_languages" directive to set which languages are considered okay for incoming mail and if the guessed language is not okay, UNWANTED_LANGUAGE_BODY is triggered. Alternatively you can use the X-Languages metadata header directly in rules.

It will always add the results to a "X-Languages" name-value pair in the message metadata data structure. This may be useful as Bayes tokens and can also be used in rules for scoring. The results can also be added to marked-up messages using "add_header", with the _LANGUAGES_ tag. See Mail::SpamAssassin::Conf for details.

Note: the language cannot always be recognized with sufficient confidence. In that case, no action is taken.

You can use _TEXTCATRESULTS_ tag to view the internal ngram-scoring, it might help fine-tuning settings.

Examples of using X-Languages header directly in rules:

header OK_LANGS X-Languages =~ /\ben\b/
score OK_LANGS -1

header BAD_LANGS X-Languages =~ /\b(?:ja|zh)\b/
score BAD_LANGS 1

USER OPTIONS

ok_languages xx [ yy zz ... ] (default: all)

This option is used to specify which languages are considered okay for incoming mail. SpamAssassin will try to detect the language used in the message body text.

Note that the language cannot always be recognized with sufficient confidence. In that case, no action is taken.

The rule UNWANTED_LANGUAGE_BODY is triggered if none of the languages detected are in the "ok" list. Note that this is the only effect of the "ok" list. It does not act as a welcomelist against any other form of spam scanning.

In your configuration, you must use the two or three letter language specifier in lowercase, not the English name for the language. You may also specify all if a desired language is not listed, or if you want to allow any language. The default setting is all.

Examples:

ok_languages all         (allow all languages)
ok_languages en          (only allow English)
ok_languages en ja zh    (allow English, Japanese, and Chinese)

Note: if there are multiple ok_languages lines, only the last one is used.

Select the languages to allow from the list below:

af - Afrikaans

am - Amharic

ar - Arabic

be - Byelorussian

bg - Bulgarian

bs - Bosnian

ca - Catalan

cs - Czech

cy - Welsh

da - Danish

de - German

el - Greek

en - English

eo - Esperanto

es - Spanish

et - Estonian

eu - Basque

fa - Persian

fi - Finnish

fr - French

fy - Frisian

ga - Irish Gaelic

gd - Scottish Gaelic

he - Hebrew

hi - Hindi

hr - Croatian

hu - Hungarian

hy - Armenian

id - Indonesian

is - Icelandic

it - Italian

ja - Japanese

ka - Georgian

ko - Korean

la - Latin

lt - Lithuanian

lv - Latvian

mr - Marathi

ms - Malay

ne - Nepali

nl - Dutch

no - Norwegian

pl - Polish

pt - Portuguese

qu - Quechua

rm - Rhaeto-Romance

ro - Romanian

ru - Russian

sa - Sanskrit

sco - Scots

sk - Slovak

sl - Slovenian

sq - Albanian

sr - Serbian

sv - Swedish

sw - Swahili

ta - Tamil

th - Thai

tl - Tagalog

tr - Turkish

uk - Ukrainian

vi - Vietnamese

yi - Yiddish

zh - Chinese (both Traditional and Simplified)

zh.big5 - Chinese (Traditional only)

zh.gb2312 - Chinese (Simplified only)

inactive_languages xx [ yy zz ... ] (default: see below)

This option is used to specify which languages will not be considered when trying to guess the language. For performance reasons, supported languages that have fewer than about 5 million speakers are disabled by default. Note that listing a language in ok_languages automatically enables it for that user.

The default setting is:

bs cy eo et eu fy ga gd is la lt lv rm sa sco sl yi

That list is Bosnian, Welsh, Esperanto, Estonian, Basque, Frisian, Irish Gaelic, Scottish Gaelic, Icelandic, Latin, Lithuanian, Latvian, Rhaeto-Romance, Sanskrit, Scots, Slovenian, and Yiddish.

textcat_max_languages N (default: 3)

The maximum number of languages any one message can simultaneously match before its classification is considered unknown. You can try reducing this to 2 or possibly even 1 for more confident results, as it's unusual for a message to contain multiple languages.

Read description for textcat_acceptable_score also, as these settings are closely related. Scoring affects how many languages might be matched and here we set the "false positive limit" where we think the engine can't decide what languages message really contain.

textcat_optimal_ngrams N (default: 0)

If the number of ngrams is lower than this number then they will be removed. This can be used to speed up the program for longer inputs. For shorter inputs, this should be set to 0.

textcat_max_ngrams N (default: 400)

The maximum number of ngrams that should be compared with each of the languages models (note that each of those models is used completely).

textcat_acceptable_score N (default: 1.02)

Include any language that scores at least textcat_acceptable_score in the returned list of languages.

This setting is basically a percentile range. Any language having internal ngram-score within N-percent of the best score is included into results. Larger values than 1.05 are not recommended as it can generate many false matches. A setting of 1.00 would mean a single best scoring language is always forcibly selected, but this is not recommended as then textcat_max_languages can't do its job classifying language as uncertain.

Read the description for textcat_max_languages, as these are settings are closely related.

You can use _TEXTCATRESULTS_ tag to view the internal ngram-scoring, it might help fine-tuning settings.