Skip to main content

A spelling and grammar corrector for Icelandic

Project description

GreynirCorrect: Spelling and grammar correction for Icelandic


GreynirCorrect is a Python 3 (>= 3.7) package and command line tool for checking and correcting spelling and grammar in Icelandic text.

GreynirCorrect relies on the Greynir package, by the same authors, to tokenize and parse text.

GreynirCorrect is documented in detail here.

The software has three main modes of operation, described below.

As a fourth alternative, you can call the JSON REST API of to apply the GreynirCorrect spelling and grammar engine to your text, as documented here.

Token-level correction

GreynirCorrect can tokenize text and return an automatically corrected token stream. This catches token-level errors, such as spelling errors and erroneous phrases, but not grammatical errors. Token-level correction is relatively fast.

Full grammar analysis

GreynirCorrect can analyze text grammatically by attempting to parse it, after token-level correction. The parsing is done according to Greynir’s context-free grammar for Icelandic, augmented with additional production rules for common grammatical errors. The analysis returns a set of annotations (errors and suggestions) that apply to spans (consecutive tokens) within sentences in the resulting token list. Full grammar analysis is considerably slower than token-level correction.

Command-line tool

GreynirCorrect can be invoked as a command-line tool to perform token-level correction and, optionally, grammar analysis. The command is correct infile.txt outfile.txt. The command-line tool is further documented below.


To perform token-level correction from Python code:

>>> from reynir_correct import tokenize
>>> g = tokenize("Af gefnu tilefni fékk fékk daninn vilja sýnum "
>>>     "framgengt í auknu mæli.")
>>> for tok in g:
>>>     print("{0:10} {1}".format(tok.txt or "", tok.error_description))


Að         Orðasambandið 'Af gefnu tilefni' var leiðrétt í 'að gefnu tilefni'
fékk       Endurtekið orð ('fékk') var fellt burt
Daninn     Orð á að byrja á hástaf: 'daninn'
vilja      Orðasambandið 'vilja sýnum framgengt' var leiðrétt í 'vilja sínum framgengt'
í          Orðasambandið 'í auknu mæli' var leiðrétt í 'í auknum mæli'

To perform full spelling and grammar analysis of a sentence from Python code:

from reynir_correct import check_single
sent = check_single("Páli, vini mínum, langaði að horfa á sjónnvarpið.")
for annotation in sent.annotations:


000-004: P_WRONG_CASE_þgf_þf Á líklega að vera 'Pál, vin minn' / [Pál , vin minn]
009-009: S004   Orðið 'sjónnvarpið' var leiðrétt í 'sjónvarpið'


'Páli, vini mínum, langaði að horfa á sjónvarpið.'

The annotation.start and annotation.end properties (here start is 0 and end is 4) contain the 0-based indices of the first and last tokens to which the annotation applies. The annotation.start_char and annotation.end_char properties contain the indices of the first and last character to which the annotation applies, within the original input string.

P_WRONG_CASE_þgf_þf and S004 are error codes.

For more detailed, low-level control, the check_errors() function supports options and can produce various types of output:

from reynir_correct import check_errors
x = "Páli, vini mínum, langaði að horfa á sjónnvarpið."
options = { "input": x, "annotations": True, "format": "text" }
s = check_errors(**options)
for i in s.split("\n"):


Pál, vin minn, langaði að horfa á sjónvarpið.
000-004: P_WRONG_CASE_þgf_þf Á líklega að vera 'Pál, vin minn' | 'Páli, vini mínum,' -> 'Pál, vin minn' | None
009-009: S004   Orðið 'sjónnvarpið' var leiðrétt í 'sjónvarpið' | 'sjónnvarpið' -> 'sjónvarpið' | None

The following options can be specified:



Default value


Defines the input. Can be a string or an iterable of strings, such as a file object.


(alias grammar)

Defines the level of correction. If False, only token-level annotation is carried out. If True, sentence-level annotation is carried out.



If True, sentences that cannot be parsed are annotated in their entirety as errors.



If True, annotations can in certain cases contain a list of possible corrections, for the user to pick from.



If True, more farfetched automatically suggested corrections are suppressed.



The value is a set of strings to whitelist. Each string is a word that should not be marked as an error or corrected. The comparison is case-sensitive.



The input contains a single sentence only. Sentence splitting should not be attempted.



A set of error codes that should be ignored in the annotation process.


An overview of error codes is available here.


GreynirCorrect runs on CPython 3.7 or newer, and on PyPy 3.7 or newer. It has been tested on Linux, macOS and Windows. The PyPi package includes binary wheels for common environments, but if the setup on your OS requires compilation from sources, you may need

$ sudo apt-get install python3-dev

…or something to similar effect to enable this.


To install this package (assuming you have Python >= 3.7 with pip installed):

$ pip install reynir-correct

If you want to be able to edit the source, do like so (assuming you have git installed):

$ git clone
$ cd GreynirCorrect
$ # [ Activate your virtualenv here if you have one ]
$ pip install -e .

The package source code is now in GreynirCorrect/src/reynir_correct.

The command line tool

After installation, the corrector can be invoked directly from the command line:

$ correct input.txt output.txt


$ echo "Þinngið samþikkti tilöguna" | correct
Þingið samþykkti tillöguna

Input and output files are encoded in UTF-8. If the files are not given explicitly, stdin and stdout are used for input and output, respectively.

Empty lines in the input are treated as sentence boundaries.

By default, the output consists of one sentence per line, where each line ends with a single newline character (ASCII LF, chr(10), "\n"). Within each line, tokens are separated by spaces.

The following (mutually exclusive) options can be specified on the command line:


Output token objects in CSV format, one per line. Sentences are separated by lines containing 0,"",""


Output token objects in JSON format, one per line.


Normalize punctuation, causing e.g. quotes to be output in Icelandic form and hyphens to be regularized.


Output whole-sentence annotations, including corrections and suggestions for spelling and grammar. Each sentence in the input is output as a text line containing a JSON object, terminated by a newline.

The CSV and JSON formats of token objects are identical to those documented for the Tokenizer package.

The JSON format of whole-sentence annotations is identical to the one documented for the HTTPS REST API.

Type correct -h to get a short help message.

Command Line Examples

$ echo "Atvinuleysi jógst um 3%" | correct
Atvinnuleysi jókst um 3%
$ echo "Barnið vil grænann lit" | correct --csv

Note how vil is not corrected, as it is a valid and common word, and the correct command does not perform grammar checking by default.

$ echo "Pakkin er fyrir hestin" | correct --json
{"k":"BEGIN SENT"}
{"k":"END SENT"}

To perform whole-sentence grammar checking and annotation as well as spell checking, use the --grammar option:

$ echo "Ég kláraði verkefnið þrátt fyrir að ég var þreittur." | correct --grammar
   "original":"Ég kláraði verkefnið þrátt fyrir að ég var þreittur.",
   "corrected":"Ég kláraði verkefnið þrátt fyrir að ég var þreyttur.",
      {"k":6,"x":"kláraði","o":" kláraði"},
      {"k":6,"x":"verkefnið","o":" verkefnið"},
      {"k":6,"x":"þrátt fyrir","o":" þrátt fyrir"},
      {"k":6,"x":"að","o":" að"},
      {"k":6,"x":"ég","o":" ég"},
      {"k":6,"x":"var","o":" var"},
      {"k":6,"x":"þreyttur","o":" þreittur"},
         "text":"Hér er réttara að nota viðtengingarhátt
            sagnarinnar 'vera', þ.e. 'væri'.",
         "detail":"Í viðurkenningarsetningum á borð við 'Z'
            í dæminu 'X gerði Y þrátt fyrir að Z' á sögnin að vera
            í viðtengingarhætti fremur en framsöguhætti.",
         "text":"Orðið 'þreittur' var leiðrétt í 'þreyttur'",

The output has been formatted for legibility - each input sentence is actually represented by a JSON object in a single line of text, terminated by newline.

Note that the corrected field only includes token-level spelling correction (in this case þreittur -> þreyttur), but no grammar corrections. The grammar corrections are found in the annotations list. To apply corrections and suggestions from the annotations, replace source text or tokens (as identified by the start and end, or start_char and end_char properties) with the suggest field, if present.


To run the built-in tests, install pytest, cd to your GreynirCorrect subdirectory (and optionally activate your virtualenv), then run:

$ python -m pytest


Parts of this software are developed under the auspices of the Icelandic Government’s 5-year Language Technology Programme for Icelandic, which is managed by Almannarómur and described here (English version here).

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

reynir-correct-3.4.6.tar.gz (3.8 MB view hashes)

Uploaded source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page