--- layout: default title: Tokenizers nav_order: 60 has_children: false has_toc: false --- # Tokenizers A tokenizer receives a stream of characters and splits the text into individual _tokens_. A token consists of a term (usually, a word) and metadata about this term. For example, a tokenizer can split text on white space so that the text `Actions speak louder than words.` becomes [`Actions`, `speak`, `louder`, `than`, `words.`]. The output of a tokenizer is a stream of tokens. Tokenizers also maintain the following metadata about tokens: - The **order** or **position** of each token: This information is used for word and phrase proximity queries. - The starting and ending positions (**offsets**) of the tokens in the text: This information is used for highlighting search terms. - The token **type**: Some tokenizers (for example, `standard`) classify tokens by type, for example, `` or ``. Simpler tokenizers (for example, `letter`) only classify tokens as type `word`. You can use tokenizers to define custom analyzers. ## Built-in tokenizers The following tables list the built-in tokenizers that OpenSearch provides. ### Word tokenizers Word tokenizers parse full text into words. Tokenizer | Description | Example :--- | :--- | :--- `standard` | - Parses strings into tokens at word boundaries
- Removes most punctuation | `It’s fun to contribute a brand-new PR or 2 to OpenSearch!`
becomes
[`It’s`, `fun`, `to`, `contribute`, `a`,`brand`, `new`, `PR`, `or`, `2`, `to`, `OpenSearch`] `letter` | - Parses strings into tokens on any non-letter character
- Removes non-letter characters | `It’s fun to contribute a brand-new PR or 2 to OpenSearch!`
becomes
[`It`, `s`, `fun`, `to`, `contribute`, `a`,`brand`, `new`, `PR`, `or`, `to`, `OpenSearch`] `lowercase` | - Parses strings into tokens on any non-letter character
- Removes non-letter characters
- Converts terms to lowercase | `It’s fun to contribute a brand-new PR or 2 to OpenSearch!`
becomes
[`it`, `s`, `fun`, `to`, `contribute`, `a`,`brand`, `new`, `pr`, `or`, `to`, `opensearch`] `whitespace` | - Parses strings into tokens at white space characters | `It’s fun to contribute a brand-new PR or 2 to OpenSearch!`
becomes
[`It’s`, `fun`, `to`, `contribute`, `a`,`brand-new`, `PR`, `or`, `2`, `to`, `OpenSearch!`] `uax_url_email` | - Similar to the standard tokenizer
- Unlike the standard tokenizer, leaves URLs and email addresses as single terms | `It’s fun to contribute a brand-new PR or 2 to OpenSearch opensearch-project@github.com!`
becomes
[`It’s`, `fun`, `to`, `contribute`, `a`,`brand`, `new`, `PR`, `or`, `2`, `to`, `OpenSearch`, `opensearch-project@github.com`] `classic` | - Parses strings into tokens on:
  - Punctuation characters that are followed by a white space character
  - Hyphens if the term does not contain numbers
- Removes punctuation
- Leaves URLs and email addresses as single terms | `Part number PA-35234, single-use product (128.32)`
becomes
[`Part`, `number`, `PA-35234`, `single`, `use`, `product`, `128.32`] `thai` | - Parses Thai text into terms | `สวัสดีและยินดีต`
becomes
[`สวัสด`, `และ`, `ยินดี`, `ต`] ### Partial word tokenizers Partial word tokenizers parse text into words and generate fragments of those words for partial word matching. Tokenizer | Description | Example :--- | :--- | :--- `ngram`| - Parses strings into words on specified characters (for example, punctuation or white space characters) and generates n-grams of each word | `My repo`
becomes
[`M`, `My`, `y`, `y `,  ,  r, `r`, `re`, `e`, `ep`, `p`, `po`, `o`]
because the default n-gram length is 1--2 characters `edge_ngram` | - Parses strings into words on specified characters (for example, punctuation or white space characters) and generates edge n-grams of each word (n-grams that start at the beginning of the word) | `My repo`
becomes
[`M`, `My`]
because the default n-gram length is 1--2 characters ### Structured text tokenizers Structured text tokenizers parse structured text, such as identifiers, email addresses, paths, or ZIP Codes. Tokenizer | Description | Example :--- | :--- | :--- `keyword` | - No-op tokenizer
- Outputs the entire string unchanged
- Can be combined with token filters, like lowercase, to normalize terms | `My repo`
becomes
`My repo` `pattern` | - Uses a regular expression pattern to parse text into terms on a word separator or to capture matching text as terms
- Uses [Java regular expressions](https://docs.oracle.com/javase/8/docs/api/java/util/regex/Pattern.html) | `https://opensearch.org/forum`
becomes
[`https`, `opensearch`, `org`, `forum`] because by default the tokenizer splits terms at word boundaries (`\W+`)
Can be configured with a regex pattern `simple_pattern` | - Uses a regular expression pattern to return matching text as terms
- Uses [Lucene regular expressions](https://lucene.apache.org/core/8_7_0/core/org/apache/lucene/util/automaton/RegExp.html)
- Faster than the `pattern` tokenizer because it uses a subset of the `pattern` tokenizer regular expressions | Returns an empty array by default
Must be configured with a pattern because the pattern defaults to an empty string `simple_pattern_split` | - Uses a regular expression pattern to split the text at matches rather than returning the matches as terms
- Uses [Lucene regular expressions](https://lucene.apache.org/core/8_7_0/core/org/apache/lucene/util/automaton/RegExp.html)
- Faster than the `pattern` tokenizer because it uses a subset of the `pattern` tokenizer regular expressions | No-op by default
Must be configured with a pattern `char_group` | - Parses on a set of configurable characters
- Faster than tokenizers that run regular expressions | No-op by default
Must be configured with a list of characters `path_hierarchy` | - Parses text on the path separator (by default, `/`) and returns a full path to each component in the tree hierarchy | `one/two/three`
becomes
[`one`, `one/two`, `one/two/three`]