Finding errors using n-gram data

LanguageTool can make use of large n-gram data sets to detect errors with words that are often confused, like their and there. The n-gram data set is huge and thus not part of the LT download. To make use of it, you have two choices:

  • Use the form on, which always has the latest and best ngram data.
  • Set up your own server with the n-gram data or use it locally in the LanguageTool stand-alone version.

To use the data locally:

  1. Make sure you have a fast disk, i.e. an SSD. Without an SSD, using this data can make LanguageTool much slower.
  2. Download the data (8GB!) from - note: data is currently only available for English, German, French, and Spanish (plus some data for untested languages).
  3. Unzip it and put it in its own directory named en, de, fr, or es, depending on the language. The path you need to set in the next step is the directory that the en etc. directory is in, not that directory itself.
  4. Then, depending on how you use LanguageTool:
    • Stand-alone user interface and LibreOffice/OpenOffice add-on: open the Options dialog and set the n-gram directory. For the stand-alone version you now need to restart LanguageTool.
    • Command line: start with the --languagemodel option pointing to the ngram-index directory.
    • Server mode: start with the --config file option. This properties file needs to have a languageModel=… entry pointing to the ngram-index directory.
  1. Test with these sentences. These are examples of errors that can only be detected using the n-gram rule, as of September 2015:
    • English:
        • I can't remember how to go their.
        • I didn't now where it came from.
        • Alabama has for of the world's largest stadiums.
    • German:
        • Während Sie das Ganze mir einem Holzlöffel oder einem Paar Stäbchen rühren…
        • In den christlichen Traditionen gibt es unterschiedliche Anleitungen zur Mediation und Kontemplation.


An n-gram is a contiguous sequence of n items from a text, like a girl (2-gram) or a tall girl (3-gram). Once you have a large amount of these n-grams with their number of occurrences, you can use this to detect errors in texts. For example, in This is there last chance to escape., LanguageTool will look at the context of there, considering up to three words:

This is there, is there last, there last chance

The probabilities of these n-grams are then compared to the probabilities of:

This is their, is their last, their last chance

If the probability of the n-grams with their is higher than of those with there, LanguageTool assumes there's an error in the input sentence. This approach is language independent, but we currently only supply the n-gram data for English and German.

We use this data set from Google, which is very similar to what Google uses for its n-gram viewer.

Here you can see the confusion pairs we support so far:

Rule Developers

How to add words pairs is documented at Adding n-gram data rules.


Technical background information can be found on this page.

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License