[Question]: How i can use Postgres with multilanguage hybrid search

[raffaeleporcellana]

Hello everyone!

I am new to LLamaIndex, I am trying to store embeddings on Postgres using PgVectorStore.
In the retrieval phase i want combine embedding search and BM25 and I see that in the creation of vector store there is the hybrid search params to set a textual search.

vector_store = PGVectorStore.from_params(
    database="vector_db",
    host="localhost",
    password="password",
    port=5432,
    user="postgres",
    table_name="{tablename}",
    embed_dim=1536,
    hybrid_search=True <----------
    text_search_config: str = "english",

)

I see that it's possible to specify also a language and the default value is english (text_search_config). But i have multiple documents in 3 different languages.
In mongo i could create a textual_index and use a language analyzer like

{
  "name":"travel",
  "mappings": {
    "dynamic": false,
    "fields": {
      "name": {
        "type": "string",
        "analyzer": "lucene.english"
      },
      "name": {
        "type": "string",
        "analyzer": "lucene.french"
      },
      "name": {
        "type": "string",
        "analyzer": "lucene.german"
      }
    }
  }
}

Is it possible something similar?

[dosubot[bot]]

Hi @raffaeleporcellana! I'm Dosu and I’m helping the llama_index team.

PGVectorStore in LlamaIndex lets you set the language for hybrid (embedding + BM25) search using the text_search_config parameter, but this setting applies to the entire table/vector store—you can't assign different analyzers per document like you can in MongoDB. If you need true multilanguage support, you'll need to create a separate table (and PGVectorStore instance) for each language, each with its own text_search_config (e.g., "english", "french", "german") source.

There’s currently no built-in support for automatic language detection or per-document language analyzers in PGVectorStore—multilingual scenarios require manual management at the table level. If you want to enable hybrid search, set hybrid_search=True and specify the language with text_search_config when you create each store source.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}