Emacs minor mode that detects the language of what you’re typing. Automatically switches the spell checker and typo-mode (if present).
Key features:
- Detection algorithm is robust, efficient, and dead simple. Based on character trigrams.
- Support for many languages. More can be easily added.
- Stays out of your way. Set up once, then forget it exists.
- Works with documents written in multiple languages.
I write a lot of text in multiple languages and was getting tired of constantly having to switch the dictionary of my spell-checker. In true Emacs spirit, I decided to dust off my grandpa’s parentheses and wrote some code to address this problem. The result is guess-language-mode, a minor mode for Emacs that guesses the language of the current paragraph and then changes the dictionary of ispell and the language settings of typo-mode (if present). It also reruns Flyspell on the current paragraph, but only on that paragraph because I want to leave paragraphs in other languages untouched. Language guessing is triggered when Flyspell detects an unknown word, but only if the paragraph has enough material to allow for robust detection of the language (~ 35 characters).
Currently, the following languages are supported: Arabic, Czech, Danish, Dutch, English, Esperanto, Finnish, French, German, Italian, Norwegian, Polish, Portuguese, Russian, Serbian (Cyrillic and Latin), Slovak, Slovenian, Spanish, Swedish.
It is easy to add more languages and this repository includes the necessary language statistics for 47 additional languages. (These were copied from guess_language.py.) If we already have the required language data (see directory trigrams), all you need to do is to add an entry to the variable guess-language-langcodes. See here for the commit that added support for Serbian. See the code of typo-mode to determine the quoting style needed for the language that you’re adding. An overview of quoting styles across languages can be found on Wikipedia. PRs adding new languages are welcome.
This mode assumes that Flyspell is activated and configured for all relevant languages, i.e., those listed in guess-language-languages. If typo-mode is present, guess-language also changes the language there. Typo-mode is not a dependency, though.
Guess-language-mode is available through MELPA.
(require 'guess-language)
;; Optionally:
(setq guess-language-languages '(en de))
(setq guess-language-min-paragraph-length 35)guess-language-languages defines the candidate languages that should be considered. It is recommended to only include languages that are actually used because this improves performance. Languages are identified using ISO 639-1 codes (see table below).
guess-language-min-paragraph-length specifies the minimal length that a paragraph needs to have before guess-language-mode starts guessing. Based on some informal tests texts shorter than 30 characters are not enough to give good results. However, above 40 characters the algorithm performs well. Of course, these numbers depend on the target language (some are easier to detect than others) and on the (number of) candidate languages that are considered. (Open the files in the directory testdata and do M-x guess-language-mark-lines to see for yourself.)
For each language, there is a default Ispell dictionary that guess-language-mode tries to use. However, for some languages there are several dictionaries available and guess-language can’t know which one you’d like to use. For example, there are several different dictionaries for German and for English. If the dictionary that guess-language-mode uses by default is not present, you will get an error message like the following:
Error in post-command-hook (flyspell-post-command-hook): (error "Undefined dictionary: en")In this case, use the variable guess-language-langcodes to tell guess-language-mode which dictionary should be used instead. For example, use the following definition if you want to use British English and Swiss German:
(setq guess-language-langcodes
'((en . ("en_GB" "English"))
(de . ("de_CH" "German"))))The key of each entry in this alist is an ISO 639-1 language code. The first element of the value is the name of the dictionary that should be used (i.e., what you would enter when setting the language via M-x ispell-change-dictionary). The second element is the name of the language setting that should be used with typo-mode (if present). If a language is not supported by typo-mode or if you are not using typo-mode, enter nil.
For a list of all dictionaries available for spell-checking, use the following:
(mapcar 'car ispell-dictionary-alist)
Languages that are currently supported by guess-language-mode:
| Language | IDO 639-1 code | Default Ispell dictionary | Default typo-mode setting |
|---|---|---|---|
| Arabic | ar | ar | |
| Czech | cs | czech | Czech |
| Danish | da | dansk | |
| Dutch | nl | nederlands | |
| English | en | en | English |
| Esperanto | eo | esperanto | English |
| Finnish | fi | finnish | Finnish |
| French | fr | francais | French |
| German | de | de | German |
| Italian | it | italiano | Italian |
| Norwegian | nb | norsk | |
| Polish | pl | polish | |
| Portuguese | pt | portuguese | |
| Russian | ru | russian | Russian |
| Serbian (Cyrillic) | sr | serbian | German (most similar to Serbian) |
| Serbian (Latin) | sr | sr-lat | German (most similar to Serbian) |
| Slovak | sk | slovak | |
| Slovenian | sl | slovenian | |
| Spanish | es | spanish | |
| Swedish | sv | svenska |
While changing the spell-checker’s dictionary is the main purpose of guess-language, there are other things that a user might want to do when a new language is detected, for instance, a user might want to change the input method. Things like that can be easily achieved by adding custom functions to the hook guess-language-after-detection-functions. Functions on this hook take three arguments:
LANG | the language that was detected |
BEGINNING | the beginning of the region in which the new language was detected |
END | the end of the region |
Template:
(defun my-custom-function (lang beginning end)
(do-something))
(add-hook 'guess-language-after-detection-functions #'my-custom-function)Activate guess-language-mode in the buffer in which you want to use it. To activate it automatically in buffers containing text (as opposed to code), add guess-language mode to text-mode-hook:
(add-hook 'text-mode-hook (lambda () (guess-language-mode 1)))The code snipped below illustrates how guess-language can be configured to automatically change the voice used by the text-to-speech engine Festival (install festival.el for this to work):
(defun guess-language-switch-festival-function (lang beginning end)
"Switch the voice used by festival.
LANG is the ISO 639-1 code of the language (as a
symbol). BEGINNING and END are the endpoints of the region in
which LANG was detected but these are ignored."
(when (and (featurep 'festival)
(festivalp))
(pcase lang
('en (festival-voice-english-female))
('de (festival-voice-german-female)))))
(add-hook 'guess-language-after-detection-functions #'guess-language-switch-festival-function)The pcase needs to be modified to use the voices that are installed on your system. Refer to the documentation of Festival for details.
Synosaurus is an Emacs package providing access to a German or English thesaurus. Using the code below the language of the thesaurus is automatically changed to the language of the current paragraph. Refer to the documentation of Synosaurus for details.
(defun guess-language-switch-synosaurus (lang beginning end)
"Switch the thesaurus language.
LANG is the ISO 639-1 code of the language (as a
symbol). BEGINNING and END are the endpoints of the region in
which LANG was detected. These are ignored."
(when (featurep 'synosaurus)
(pcase lang
('en (setq synosaurus-backend 'synosaurus-backend-wordnet))
('de (setq synosaurus-backend 'synosaurus-backend-openthesaurus)))))
(add-hook 'guess-language-after-detection-functions #'guess-language-switch-synosaurus)- Support for Latin Serbian is based on trigrams transliterated from Cyrillic Serbian. Since some Cyrillic trigrams transliterate to 4-grams in Latin, we truncated those but as a result have two duplicates, “e n” and “ra “. Not ideal but the results are probably still robust enough. Nonetheless, it would be good if someone could compute proper Latin trigrams one day.