For AI agents: this page is also available as Markdown at /docs/snowball-stemming.md. Every docs page is: add .md to its URL. llms.txt lists them all.

Snowball stemming (European languages)

Stemming reduces words to a common stem, so running matches run and German Häuser matches Haus. The snowball filter stems with the Snowball algorithms. Each language also has a ready-made snowball_<language> analyzer, such as snowball_german or snowball_english:

{"fields": {"title_de": {"type": "text", "analyzer": "snowball_german"}}}

The analyzers run unicode_word, nfkc_cf, the language's stop words and snowball.

#Languages

arabic, armenian, basque, catalan, danish, dutch, dutch_porter, english (Porter2), esperanto, estonian, finnish, french, german, greek, hindi, hungarian, indonesian, irish, italian, lithuanian, nepali, norwegian, porter (the original English Porter stemmer), portuguese, romanian, russian, serbian, spanish, swedish, tamil, turkish and yiddish.

esperanto, nepali, serbian, tamil and yiddish have no stop list, so their analyzers only stem. For Arabic, the arabic analyzer also normalizes spelling and usually works better than snowball_arabic.

#In your own chain

Set the filter's required language parameter:

{
  "type": "text",
  "analyzer": {
    "tokenizer": "unicode_word",
    "filters": ["nfkc_cf", {"name": "snowball", "params": {"language": "french"}}]
  }
}

• Put nfkc_cf or lowercase before snowball: it expects lowercase input.
• Leave fold out for languages whose stemmer reads accents, such as German or French, unless you want accented and unaccented forms to match.
• Prefix, wildcard and fuzzy terms are not stemmed: häuser* matches terms that start with häuser, and the index holds haus.

#English

For English, the _t suffix's KStem stemming is gentler than Porter2: snowball_english matches more word forms, and also more unrelated words. Use _t unless recall matters more than precision.

#Through the compatible APIs

Elasticsearch and OpenSearch language analyzers such as german or french, the snowball and stemmer filters, and Solr's text_<language> types and SnowballPorterFilterFactory map to these components. Elasticsearch's light and minimal stemmers use the Snowball stemmer of the language, with a warning. See Through the compatible APIs.

View this page as Markdown