Install and run LanguageTool on a local machine

Download ngram

wget https://languagetool.org/download/ngram-data/ngrams-en-20150817.zip
wget https://languagetool.org/download/ngram-data/ngrams-de-20150819.zip
unzip *.zip
rm *.zip

Download word2vec models

wget https://languagetool.org/download/word2vec/de.zip
wget https://languagetool.org/download/word2vec/en.zip
unzip *.zip
rm *.zip

In case you are interested in how neural networks were trained using the word2vec models, the source code can be found here on Github: gulp21/languagetool-neural-network.

Use fasttext for language detection (optional)

Install fasttext

see https://fasttext.cc for more information about fasttext.

git clone https://github.com/facebookresearch/fasttext.git
cd fasttext
make

Download fasttext data

mkdir fastTextData; cd !$
## wget https://dl.fbaipublicfiles.com/fasttext/vectors-english/wiki-news-300d-1M-subword.vec.zip
wget https://dl.fbaipublicfiles.com/fasttext/supervised-models/lid.176.bin

Test it!

java -jar LanguageTool-4.9.1-stable/languagetool-commandline.jar --languagemodel ngram-data --word2vecmodel word2vec -l en-GB test.txt

A total of 14 potential issues was reported. I manually checked the results with the example provided on the website of the tool, and the results are consistent.

I experimented a bit, and observed the following:

Using nothing than -l en (without the country code GB) reported 8 issues
Using -l en-GB reported 10 issues.
Using --languagemodel alone did not increase the number of issues reported. Is it automatically activated?
Using --word2vecmodel increased the number of issues to 14. However, the additional reported issues are mostly redundant.

On a regular laptop (HP 9470m, i5-3427U, 8G memory), it took about 6.5s to examine the ext, or about 1.2 second to examine one sentence. The speed may not be the fastest, but it is much faster than I can write:)

The whole directory is about 21G, including ngram and word2vec models for English and German. It will probably fit on a 32GB SD card.

Set up the local web server

See the tutorial to run a HTTP server of LanguageTools.

java -cp /opt/LanguageTool/LanguageTool-4.9.1-stable/languagetool-server.jar org.languagetool.server.HTTPServer --port 8081 --allow-origin "*" --public --config /opt/LanguageTool/config/server-config

The files used by LanguageTool is specified in the config/server-config file. The port is set by the --port option as 8081. The --alow-origin parameter allows user to use explorer plugins (which I am not using though). The parameter --public makes sure that we can visit the server from other machines.

Failed attemp to install the system on a Raspberry pi.

I wanted to migrate the system onto a Raspberry Pi and run a server that checks texts for me. However, it failed because the default linux build compaints about architecture incompability of boost libraries. When I tried to compile from the source code, maven failed to find the dependency com.google.protobuf:protoc:exe:linux-arm_32:3.11.0. Notice that RaspberriPi runs on ARM chips. Given that I could not find the dependency, I decide to stop there.

In case anyone has successfully compiled LanguageTools on RaspberriPi systems, please kindly let me know.

To do

Configure the tool to check markdown files

Accio/2020-05-InstallAndRunLanguageTools