Query spell checker

Amazon Kendra Spell Checker suggests spell corrections for a query. This can help you keep occurrences of zero search results to a minimum and return relevant results. Your users might receive zero search results from misspelled queries with no matching results or no returned documents. Or, your users might receive irrelevant search results from misspelled queries.

Spell Checker is designed to suggest corrections for misspelled words based on words that appear in your indexed documents and how closely a corrected word matches a misspelled word. For example, if the word 'statements' appears in your indexed documents, then this could closely match the misspelled word 'statments' in the query 'year-end financial statments'.

Spell Checker returns the intended or corrected words that replace misspelled words in the original query text. For example, 'depoying kendre search' could return 'deploying Kendra search' You can also use offset locations provided in the API to highlight or italicize the returned corrected words in a query in your front-end application. In the console, the corrected words are highlighted or italicized by default. For example, 'deploying Kendra search'.

For business-specific or specialized terms that appear in your indexed documents, Spell Checker does not misunderstand these terms as spellings mistakes in the query. For example, 'amazon macie' is not corrected to 'amazon mace'.

For hyphenated words, such as 'year-end', Spell Checker treats these as individual words to suggest corrections for these words. For example, the suggested correction for 'yaer-end' could be 'year-end'.

For DOCUMENT and QUESTION_ANSWER query response types, Spell Checker suggests corrections to misspelled words based on words in the document body. The document body is more reliable than the title for suggesting corrections that closely match the misspelled words. For ANSWER query response types, Spell Checker suggests corrections based on words in the default question and answer document in your index.

You can activate Spell Checker using the SpellCorrectionConfiguration object. You set IncludeQuerySpellCheckSuggestions to TRUE. Spell Checker is activated by default in the console. It is built into the console by default.

Spell Checker can also suggest spell corrections for queries in multiple languages, not only English. For a list of languages supported for Spell Checker, see Amazon Kendra supported languages.

Using the query spell checker with default limits

Spell Checker is designed with certain defaults or limits. The following is a list of current limits that apply when you activate spell correction suggestions.

• Suggested spell corrections cannot be returned for words that are less than three characters or more than 30 characters in length. To allow for more than 30 characters or less than three characters, contact Support.

• Suggested spell corrections cannot restrict suggestions based on user access control or your access control list for user context filtering. Spell corrections are based on all words in your indexed documents, whether the words are restricted to certain users or not. If you want to avoid certain words appearing in the suggested spell corrections for queries, then do not activate SpellCorrectionConfiguration.

• Suggested spell corrections cannot be returned for words that include numbers. For example, 'how 2 not br8k ubun2'.

• Suggested spell corrections cannot use words that don't appear in your indexed documents.

• Suggested spell corrections cannot use words that are frequented less than 0.01 percent in your indexed documents. To change the 0.01% threshold, contact Support.