[Bug]: bm25 retrieve - NATURAL LANGUAGE MODE doesn't work

[shrmke]

Is there an existing issue for the same bug?

• I have checked the existing issues.

Branch Name

main

Commit ID

dd9564b

Other Environment Information

- Hardware parameters:
- OS type:
- Others:

Actual Behavior

1. create a table;
   sql:create table bm25_results(id INT PRIMARY KEY,title VARCHAR(200),content TEXT, created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP);
2. create fulltext index;
   sql:CREATE FULLTEXT INDEX ftidx_content ON bm25_results;
3. insert data；
   sql：INSERT INTO bm25_results (id, title, content) VALUES
   -> (1, '人工智能简介', '人工智能是计算机科学的一个分支...'),
   -> (2, '数据库原理', '数据库是组织和存储数据的系统...'),
   -> (3, '机器学习', '机器学习是人工智能的一个子领域...'),
   -> (4, '深度学习', '深度学习是机器学习的一个分支...');
   4.retrieve in default mode;
   sql:SELECT * FROM bm25_results WHERE MATCH(title, content) AGAINST('机器学习');
   +------+--------------+--------------------------------------------------+---------------------+
   | id | title | content | created_at |
   +------+--------------+--------------------------------------------------+---------------------+
   | 3 | 机器学习 | 机器学习是人工智能的一个子领域... | 2025-12-26 10:14:36 |
   | 4 | 深度学习 | 深度学习是机器学习的一个分支... | 2025-12-26 10:14:36 |
   +------+--------------+--------------------------------------------------+---------------------+
   2 rows in set (0.048 sec)
4. retrieve in natural language mode;
   sql:SELECT * FROM bm25_results WHERE MATCH(title, content) AGAINST('机器学习是啥'IN NATURAL LANGUAGE MODE);
   Empty set (0.010 sec)

Expected Behavior

supposed to return
+------+--------------+--------------------------------------------------+---------------------+
| id | title | content | created_at |
+------+--------------+--------------------------------------------------+---------------------+
| 3 | 机器学习 | 机器学习是人工智能的一个子领域... | 2025-12-26 10:14:36 |
| 4 | 深度学习 | 深度学习是机器学习的一个分支... | 2025-12-26 10:14:36 |
+------+--------------+--------------------------------------------------+---------------------+

Steps to Reproduce

-

Additional information

No response

[aunjgr]

改成

SELECT * FROM bm25_results WHERE MATCH(title, content) AGAINST('机器学习 是啥'IN BOOLEAN MODE);

可以得到结果

[aunjgr]

目前的行为，手动分词的话，是OR模式，搜到任意一个词就算命中，否则是精确匹配。由于语料里没有“是啥”，所以没有结果。