Lindera - ParadeDB

The Lindera tokenizer is a more advanced CJK tokenizer that uses prebuilt Chinese, Japanese, or Korean dictionaries to break text into meaningful tokens (words or phrases) rather than on individual characters. Chinese Lindera uses the CC-CEDICT dictionary, Korean Lindera uses the KoDic dictionary, and Japanese Lindera uses the IPADIC dictionary. By default, non-CJK text is lowercased, and punctuation is not ignored. As of version 0.22.4, whitespace is removed by default. On earlier versions it is preserved.

CREATE INDEX search_idx ON mock_items
USING bm25 (id, (description::pdb.lindera(chinese)))
WITH (key_field='id');

CREATE INDEX search_idx ON mock_items
USING bm25 (id, (description::pdb.lindera(korean)))
WITH (key_field='id');

CREATE INDEX search_idx ON mock_items
USING bm25 (id, (description::pdb.lindera(japanese)))
WITH (key_field='id');

To get a feel for this tokenizer, run the following command and replace the text with your own:

SELECT 'Hello world! 你好!'::pdb.lindera(chinese)::text[];

Expected Response

              text
------------------------
 {hello,world,!,你好,!}
(1 row)

Keep Whitespace

By default, whitespace is not tokenized. To include it, set keep_whitespace to true.

SELECT 'Hello world! 你好!'::pdb.lindera(chinese, 'keep_whitespace=true')::text[];

Expected Response

              text
--------------------------------
 {hello," ",world,!," ",你好,!}
(1 row)

NFKC Normalization

Set nfkc to true to apply Unicode NFKC normalization to the text before it is segmented. This collapses compatibility characters such as full-width Latin letters and digits to their canonical half-width forms, so that, for example, ＡＢＣ and ABC produce the same tokens.

SELECT 'ＡＢＣ１２３'::pdb.lindera(japanese, 'nfkc=true', 'lowercase=false')::text[];

Expected Response

   text
----------
 {ABC,123}
(1 row)

Reading Form

Set reading_form to true to replace each token with its dictionary reading form after segmentation. For Japanese (IPADIC), the surface form is replaced with its katakana reading; for Korean (KoDic), Hanja are replaced with their Hangul reading. This is supported for Japanese and Korean only.

SELECT '日本語'::pdb.lindera(japanese, 'reading_form=true', 'lowercase=false')::text[];

Expected Response

    text
------------
 {ニホンゴ}
(1 row)

​Keep Whitespace

​NFKC Normalization

​Reading Form

Keep Whitespace

NFKC Normalization

Reading Form