Querying references inside a Portable text field in Sanity

I understand you're trying to project into a Portable Text field that uses field-level translation, where you need to access references (like image assets) within the translated content blocks.

The key is to use the array projection operator [] immediately after your coalesce expression. This tells GROQ to iterate over each element in the array and apply the projection to each block:

"content": coalesce(content[$lang], content.en)[] {
  ...,
  _type == "pageImage" => {
    ...,
    "alt": coalesce(alt[$lang], alt.en),
    "imageBlock": imageBlock {
      "asset": asset->url,
      "dimensions": asset->metadata.dimensions
    }
  }
}

The [] operator is what enables you to then use conditional projections (the => operator) and dereference operations (->) within each block.

Here's a more complete example that handles different block types:

"content": coalesce(content[$lang], content.en)[] {
  _type != "pageImage" => @,
  _type == "pageImage" => {
    _type,
    _key,
    "alt": coalesce(alt[$lang], alt.en),
    "imageBlock": imageBlock {
      "asset": asset->url,
      "dimensions": asset->metadata.dimensions
    }
  }
}

In this pattern:

• The @ operator (identity operator from GROQ's pipeline components) passes through non-pageImage blocks unchanged
• pageImage blocks get the special projection with language fallback and reference dereferencing
• The asset->url syntax works because you're inside the projection context where the reference operator can resolve

The array projection operator [] is part of GROQ's pipeline architecture, where operations are chained together to transform data. The coalesce() function returns the first non-null value, and when followed by [], that value (which should be an array) is projected element-by-element.

Important note: While this approach works, the Sanity documentation on localization recommends against using field-level translation for Portable Text fields, as it can quickly consume attribute limits and create complex data structures. Document-level localization is generally preferred for Portable Text content.