Autocorrect

Autocorrect is a feature that resembles spell checking. It means that the user input, i.e. the questions that people ask, are now automatically spell checked, and they're automatically corrected whenever possible.

The Autocorrect feature checks the words that are in the user input against the following dictionaries: 

• A generic language specific library

• A custom project specific library

The project specific library is automatically updated in the CMS with any change that is made to a Q&A. The dictionary is therefore based on all the words from all the Questions and Answers in the project, thus providing a broad coverage of word use beyond the more generic language library. 

Do note that the Autocorrect tries to correct misspelled words, but cannot always succeed. The rules are set quite strictly so if a word is in no way similar to a known word from the libraries, it is just left alone. And of course Autocorrect can also make mistakes where a proper word is changed into another word. This can happen if the 'proper word' isn't available in both of the libraries and will therefore match on a word that is in there and resembles it. The solution to such exceptions is of course to add that word to a Q or A somewhere and it will be left alone, since that word is now correct.

In the Entity editor, all manual additions are automatically normalized. This includes the lemmatization process but not the autocorrect functionality.

How Will Autocorrect Impact Your Project?

For new projects: The feature will be automatically available for all new projects created in Conversational AI Cloud. Do keep in mind that some words will still need to be added to the entities, that is if they're not automatically caught and replaced by the Autocorrect logic. 

For existing projects: everything will keep working since all the misspelled words in the entities will also be in the initial dictionary. However, it's best to remove as many misspellings as possible before we implement Autocorrect in your project in order to keep your entities clean and optimize your maintenance work.

A tool has been built to check for misspellings that can be removed from the entities. Support will use this tool on your existing project to fix the misspellings via import towards the new situation.

How Will Autocorrect Impact the Words Missing From Entity Recognition?

Before the Autocorrect functionality is enabled for a project, any words containing typos usually appear on the words missing from entity recognition list. After Autocorrect is enabled, here's what happens:

• If a word containing typos is auto-corrected to the correct word, it won’t show up in the words missing from entity recognition because it's successfully matched to a entity.

• If a word is auto-corrected to the wrong word (example: the misspelled word “abcense” is auto-corrected to “absence” which is in your dictionary the word “absence”  would show up in your words missing from entity recognition list.

In a nutshell, it’s always the “result” that matters, and never the word as it has been misspelled in the user input.