LanguageTool

LanguageTool is an Open Source proofreading software for English, French, German, Polish, Russian, and more than 20 other languages. It finds many errors that a simple spell checker cannot detect.

The source repository can be found here.

About this image:

It uses the private fork source
It builds a container image from the source.
It uses a patched version of languagetool 6.5 snapshot with Lucene 8.11.3.
It uses the OpenJDK Java 21 slim base image
includes fasttext
It uses gosu
- container starts as root and executes languagetool as restricted user using exec gosu
- container fixes folder ownership for ngrams and fasttext folders
Entrypoint uses tini to suppress the container exiting with status code 143 (LanguageTool does not handle SIGTERM as it should)
optional: download ngram language modules if configured (when there have never been downloaded yet.)
- It uses the ngram module upgraded to Lucene 8 format
optional: download fasttext module (if it doesn't already exist)
optional: user mapping (make sure to check MAP_UID and MAP_GID below)
optional: set log level

Compose configuration

Please refer compose.yml configuration file.

---
services:
  languagetool:
    build: .
    container_name: languagetool
    restart: always
    cap_drop:
      - ALL
    cap_add:
      - CAP_SETUID
      - CAP_SETGID
      - CAP_CHOWN
    security_opt:
      - no-new-privileges
    ports:
      - 8010:8010
    environment:
      download_ngrams_for_langs: en
      langtool_languageModel: /ngrams
      langtool_fasttextModel: /fasttext/lid.176.bin
    volumes:
      - ./ngrams:/ngrams
      - ./fasttext:/fasttext

An example compose file can be downloaded from here.

Compose execution

You can use containerd's nerdctl, rancher desktop or docker CLI command.

git clone https://github.com/miurahr/docker-languagetool.git
cd docker-languagetool
nerdctl compose build
nerdctl compose up

Parameters

The environment parameters are split into two halves, separated by an equal, the left-hand side represents the variable names (use them as is) the right-hand side the value (change if necessary).

ENV	DEFAULT	DESCRIPTION
download_ngrams_for_langs	none	Optional: Comma separated list of languages to download ngrams for. Skips download if the ngrams for that language already exist. Valid languages: `en`, `de`, `es`, `fr` and `nl`. Example value: `en,de`
langtool_languageModel	/ngrams	Optional: The base path to the ngrams models.
langtool_fasttextBinary	/usr/local/bin/fasttext	Optional: Path to the fasttext binary. Change only if you want to test your own compiled binary. Don't forget to map it into the container as volume.
langtool_fasttextModel		Optional: The container path to the fasttext model binary. If the variable is set, the fasttext model will be downloaded if doesn't exist yet.
langtool_*		Optional: An arbitrary languagetool configuration, consisting of the prefix `langtool_` and the key name as written in the config file. Execute `docker run -ti --rm meyay/languagetool help` to see the list of config options
JAVA_XMS	256m	Optional: Minimum size of the Java heap space. Valid suffixes are `m` for megabytes and `g` for gigabytes.
JAVA_XMX	1024m	Optional: Maximum size of the Java heap space. Valid suffixes are `m` for megabytes and `g` for gigabytes. Set a higher value if you experience OOM kills, but do not use more than 1/4 of the host memory!
JAVA_GC	SerialGC	Optional: Configure the garbage collector the JVM will use. Valid options are: `SerialGC`, `ParallelGC`, `ParNewGC`, `G1GC`, `ZGC`
JAVA_OPTS		Optional: Set you own custom Java options for the JVM. This will render the other JAVA_* options useless.
MAP_UID	783	Optional: UID of the user inside the container that runs LanguageTool. If you encounter permission problems with your volumes, make sure to set the parameter to the UID of the host folder owner.
MAP_GID	783	Optional: GID of the user inside the container that runs LanguageTool. If you encounter permission problems with your volumes, make sure to set the parameter to the GID of the host folder owner.
LOG_LEVEL	INFO	Optional: set log level for LanguageTool. Valid options are: `TRACE`, `DEBUG`, `INFO`, `WARN`, `ERROR`.

Fasttext support

Now that fasttext is available since Alpine 3.19, the image switched to using the Alpine package, instead of compiling the binaries from the sources. This hopefully fixes the compatibility issue users with older cpus experienced with my previous images, that were build on a amd64v3 architecture cpu, which compiled the fasttext binary with cpu optimizations older cpus do not support.

If the debian fasttext package does not work for you, you can build a custom image to compile the fasttext binary using cpu optimizations your cpu (as long as it's x86_64 based) actually understands:

git clone https://github.com/miurahr/docker-languagetool.git
cd docker-languagetool
nerdctl build -t miurahr/languagetool:latest -f Dockerfile.fasttext .

Once the image is build, you can nerdctl compose up -d.

NOTE1: Synology users can find a git package in the SynoCommunity repository.

Changelog