ltapiserv-rs: A Rust repository from cpg314

This provides a lightweight alternative backend implementation of the LanguageTool API for offline grammar and spell checking, based on:

nlprule for grammar and style checking, using the LanguageTool rules.
symspell for spell-checking.

A simple command-line client, ltapi-client, is also provided, displaying results graphically with miette.

Quick start

Install the server, which is a single binary with no dependencies, which can be run as a systemd service. Debian/Ubuntu and Arch packages are provided. Alternatively, a Docker image is available.
Configure your client (browser, editor, ltapi-client, ...) to use the local server.

Screenshots

Using the ltapiserv-rs server with the official LanguageTool browser extension.

Output from the client CLI

Background

LanguageTool

LanguageTool is an open-source alternative to Grammarly for natural language linting (spelling, grammar, style), with a large set of rules. Multiple clients exist for its API, bringing functionalities to Firefox, Chrome, LibreOffice, Thunderbird, emacs, and many more.

Self-hosting LanguageTool

While most users access LanguageTool through the official hosted server (with a free or paid plan), the Java API server can be hosted locally, which can be particularly desirable for privacy reasons (e.g. when editing confidential documents).

Even though the browser extensions are unfortunately closed-source, they still allow custom servers to be specified.

Lightweight LanguageTool API server

Benjamin Minixhofer wrote a Rust crate, nlprule, that is able to parse and then apply LanguageTool rules noticeably faster than the original Java implementation (see this benchmark). More complex rules written in Java are not supported and spellchecking is not implemented, but nevertheless roughly 85% of the LanguageTool grammar rules (as of 2021) are available.

Using nlprule and symspell (for spell-checking), we can implement a simple LanguageTool API server in Rust that can then be called from a variety of contexts using LanguageTool clients.

The code and binaries can be found on https://github.com/cpg314/ltapiserv-rs.
See the README there for the configuration as a systemd service as well as the setup of the clients.

Comparison with the Java server

Running H.G. Wells' War of the Worlds (~6k lines and 62k words) through the two servers, using hyperfine and httpie, we get:

$ docker pull erikvl87/languagetool
$ docker run --rm -p 8010:8010 erikvl87/languagetool
  http://localhost:{port}/v2/check language=en-us text=@wells.txt'
$ for port in 8875 8010; do http --form  POST http://localhost:$port/v2/check \
                            language=en-us text=@wells.txt | jq ".matches|length"; done
1490
1045
$ hyperfine -L port 8875,8010 --runs 10 'http --ignore-stdin --meta --form  POST \
                   http://localhost:{port}/v2/check language=en-us text=@wells.txt'

The additional false positives in ltapiserv-rs seem to come mostly from the spell-checking.

Command	Mean [s]	Min [s]	Max [s]	Relative
`ltapiserv-rs`	16.002 ± 0.629	15.566	17.745	1.00
`java`	30.594 ± 2.372	29.569	37.296	1.91 ± 0.17

With only a paragraph (to simulate something close to the normal use of LanguageTool, say in emails):

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
`ltapiserv-rs`	379.7 ± 9.3	362.6	393.4	1.00

Installation

Any of the following methods will make a server available at http://localhost:8875

By default, the custom dictionary is located in ~/.local/share/ltapiserv-rs/dictionary.txt. A different path can be passed via the --dictionary option. The contents are automatically reloaded on file change.

Docker

$ docker run -d --name ltapiserv-rs -p 8875:8875 -v ~/.local/share/ltapiserv-rs:/data ghcr.io/cpg314/ltapiserv-rs:0.2.2
$ docker logs -f ltapiserv-rs

Debian/Ubuntu and Arch packages

Packages are available from the releases page, containing the server and ltapi-client. They will install a systemd service definition in /usr/lib/systemd/user/ltapiserv-rs.service, which can be enabled with:

$ systemctl --user enable --now ltapiserv-rs
$ # Check status
$ systemctl --user status ltapiserv-rs
$ # Check logs
$ journalctl --user -u ltapiserv-rs -f
$ # After updating:
$ systemctl --user restart ltapiserv-rs

tar.gz archive

For other distributions, standalone binaries are also available from the releases page.

$ sudo cp ltapiserv-rs /usr/local/bin
$ sudo chmod +x /usr/local/bin/ltapiserv-rs
$ ln -s $(pwd)/ltapiserv-rs.service ~/.config/systemd/user/ltapiserv-rs.service
$ systemctl --user daemon-reload && systemctl --user enable --now ltapiserv-rs
$ systemctl --user status ltapiserv-rs
$ # After updating:
$ systemctl --user restart ltapiserv-rs

See the above remark about the custom dictionary.

Usage / clients

The following clients have been tested. The server should be compatible with others, but there might be idiosyncrasies; don't hesitate to send a PR.

Browser extensions

Install the official LanguageTool browser extension (e.g. for Chrome or Firefox) and configure it to use your local server:

Command line client

A command line client, ltapi-client, is also included in this project.

Run text through a LanguageTool server and display the results

Usage: ltapi-client [OPTIONS] --server <SERVER> [FILENAME]

Arguments:
  [FILENAME]  Filename; if not provided, will read from stdin

Options:
  -l, --language <LANGUAGE>        [default: en-US]
  -s, --server <SERVER>            Server base URL [env: LTAPI_SERVER=http://localhost:8875]
      --json                       JSON output
      --suggestions <SUGGESTIONS>  Number of suggestions to display [default: 3]
      --pandoc                     Convert to plaintext with pandoc, removing code blocks. Line numbers are not preserved.
  -h, --help                       Print help

The return code will be 1 if any error is detected. The server address can be configured through the LTAPI_SERVER environment variable.
If pandoc is installed, the client can use it to convert input files into plain text.

The client uses miette to get a nice graphical reporting of the errors:

Example usage

$ export LTAPI_SERVER=http://localhost:8875
$ cat text.txt | ltapi-client
$ ltapi-client test.txt
$ ltapi-client --pandoc test.md

flycheck-languagetool (emacs)

See https://github.com/emacs-languagetool/flycheck-languagetool

(use-package flycheck-languagetool
  :ensure t
  :hook ((text-mode gfm-mode markdown-mode) . flycheck-languagetool-setup)
  :init
  (setq flycheck-languagetool-url "http://127.0.0.1:8875")
  :custom
  (flycheck-languagetool-active-modes '(text-mode gfm-mode markdown-mode))
  )

ltex-ls (language server protocol for markup)

See https://github.com/valentjn/ltex-ls.

This currently requires this patch to send the proper content type in the requests (this also could be done in ltapiserv-rs with an axum middleware to edit the content type).

Use the ltex.languageToolHttpServerUri variable to set the URL, e.g. with lsp-ltex in emacs:

(use-package lsp-ltex
  :ensure t
  :hook (text-mode . (lambda ()
                       (require 'lsp-ltex)
                       (lsp)))  ; or lsp-deferred
  :init
  (setq lsp-ltex-version "16.0.0"
        lsp-ltex-languagetool-http-server-uri "http://localhost:8875"
        )
)

Tools based on `languagetools-rust`

Unfortunately, tools such as cargo-languagetool and languagetool-code-comments, based on the languagetool-rust client, are for now not compatible with this server. There are two reasons:

The queries are sent as URL parameters rather than as form data. Even though the former matches the API specifications, the latter is also supported by the official server.
The client expect all fields to be contained in the response, while we only send a subset.

The solution to the first issue is simple (support query parameters depending on the Content-Type header and/or request contents). For the second, one should either expand the messages defined here, or replace them by the languagetools-rust ones (which might cause serialization issues in the form-data path), or add conversions.

Implementation details

API endpoint

The LanguageTool API is documented here. It suffices to implement the HTTP POST /v2/check endpoint that processes

pub struct Request {
    text: Option<String>,
    data: Option<String>,
    language: String,
}

and returns

pub struct Response {
    pub matches: Vec<Match>,
    pub language: LanguageResponse,
}
pub struct Match {
    pub message: String,
    pub short_message: String,
    pub offset: usize,
    pub length: usize,
    pub replacements: Vec<Replacement>,
    pub sentence: String,
    pub context_for_sure_match: usize,
    pub ignore_for_incomplete_sentence: bool,
    pub r#type: MatchType,
    pub rule: Rule,
}

The most important fields in Response are offset, length (defining the span of the suggestion), message, replacements, and Rule.

There are a couple of small tricks required to get the closed-source browser extensions to behave as expected, e.g. in displaying grammar and spelling errors with the right colours and showing tooltips.

Grammar, spelling, and repetition checkers

The main functionality, returning suggestions based on an input text, can be reduced to the following method:

pub fn suggest(&self, text: &str) -> Vec<api::Match> {
    let mut suggestions = Vec::new();
    for sentence in self.tokenizer.pipe(text) {
        debug!("Processing sentence {:#?}", sentence);
        // Grammar suggestions from nlprule
        suggestions
            .extend(self.rules.apply(&sentence).into_iter().map(Match::from));
        // Spelling and repetitions, processing the sentence token by token.
        let tokens = sentence.tokens();
        for (i, token) in tokens.iter().enumerate() {
            // ...
        }
    }
    suggestions
}

The Match::from method performs conversion between an nlprule::Suggestion to a Match, essentially copying over the span and the message.

The nlprule crate does not yet support spell checking, but we can add a basic version using the symspell crate and leveraging the tokenization we already have from nlprule. Similarly, the tokenization allows us to implement a word repetition rule that did not seem present in nlprule.

Compile from source

Binaries can also be built from source as follows:

$ cargo make build

Future work

LanguageTool (even the original implementation with all rules) seems to be failing to identify more subtle grammatical errors:

"No one would have believed in the last years of the nineteenth century that this world were being watched keenly and closely by intelligences greater than man's"

"With infinite complacency men went to and fro over this globe about his little affairs, serene in their assurance of their empire over matter."

It would be interesting to understand what the state of the art is (under a fast processing constraint).
Support more languages. German is already supported in nlprule, but adding more languages is actually non-trivial because of language-specific assumptions, see this issue and this one.
Support addition and deletion of words to the dictionary. This is pretty simple and corresponds to the /words/add and /words/delete API endpoints. However, the browser extension seems to store the dictionary locally, unless one logs in to LanguageTool Premium.
Reduce the number of false positives of the spellchecker.
Expand tests

cpg314/ltapiserv-rs