Skip to contents

Semantic embeddings for every unique word item definition, computed with Google's multilingual gemini-embedding-001 model (768 dimensions). All languages share one embedding space, so cosine similarities are meaningful both within and across languages.

Usage

get_embeddings(language = NULL, version = "current")

Arguments

language

An optional string specifying which language's embeddings to retrieve.

version

A string specifying which version of the Wordbank dataset to use, e.g. "v1.2" to pin a released version for reproducibility. Defaults to "current", the most recent release.

Value

A data frame with one row per unique item definition: language, item_definition, embedding (a list-column of numeric vectors), and dataset_version.

Examples

if (FALSE) { # \dontrun{
danish_embeddings <- get_embeddings(language = "Danish")
} # }