Multilingual discovery
Multilingual scholarly discovery

Search in your language. Discover research across languages.

Scholarly knowledge is published in many languages. Yet relevant work can remain invisible when discovery depends on exact words. This guided pilot compares keyword, semantic and hybrid retrieval over a LA Referencia subset.

Understand the idea Explore the live queries
search.lareferencia.info
远程教育 distance education
Keyword No literal matches
Semantic
Educación a distancia vía videoconferencia
Ensino remoto em debate
Gestión de la tutoría telemática…
Hybrid
Educación a distancia vía videoconferencia
Ensino remoto em debate
Enseñanza en la virtualidad…

Knowledge is multilingual. Discovery often is not.

The barrier is not research quality. It is visibility: useful work can be missed simply because the query and the record use different words or languages.

Exact-word bias

Keyword matching is literal. It can miss related work expressed with different terminology, regional phrasing or discipline-specific language.

Language silos

Equivalent ideas become disconnected when a search only retrieves records that share the language and wording of the original query.

Uneven visibility

When discovery tools favour dominant languages, relevant scholarship in other languages becomes harder for global audiences to find and use.

Search by meaning, not only by words.

Semantic multilingual search compares meaning across languages and phrasing. It complements keyword search: it does not translate documents, and proximity of meaning does not guarantee relevance.

01

Keyword Search

Finds literal terms in indexed metadata. It is useful for precise known terms, especially when the query and the record share vocabulary and language.

02

Semantic Search

Retrieves nearby meanings within a shared multilingual space, even when the query and the record use different languages or phrasing.

03

Hybrid Search

Combines literal and semantic rankings. It can preserve exact-term precision while adding meaning-based discovery.

30,063 records in the pilot subset
A bounded experiment

This is a pilot over an indexed subset, not an evaluation of the complete LA Referencia network. Deposita is used as a thematic control set: it confirms that each selected topic is represented before multilingual queries are tested without repository filters.

Four topics, twenty-four live queries.

Each topic starts with an original Portuguese phrase verified in Deposita. Every multilingual card opens the unfiltered three-column comparison over the pilot subset.

Try simple English phrasing.

Semantic search can also reduce exact-word bias within one language. These simple alternatives use the same editorial review as the multilingual examples.

Interpret the pilot with care.

Editorial review makes the comparison auditable without presenting an exploratory prototype as a comprehensive benchmark.

How the scores were produced

Subset composition

    Deposita selects the topics; it does not filter the public demonstration links.

    Credits

    Built from shared work.