By default, Docling does not automatically detect the language. It uses a fixed list of languages passed to Tesseract, with the default being ["fra", "deu", "spa", "eng"] (French, German, Spanish, English).

However, optional auto-detection is available: setting lang=["auto"] causes Docling to use Tesseract's OSD (Orientation and Script Detection), which detects the document's script and maps it to a language via the map_tesseract_script() function. No external libraries (e.g., langdetect, fasttext) are used — everything depends solely on Tesseract.

Example – enabling auto-detection:

from docling.datamodel.pipeline_options import PdfPipelineOptions, TesseractCliOcrOptions

ocr_options = TesseractCliOcrOptions(lang=["auto"])
pipeline_options = PdfPipelineOptions(
    do_ocr=True,
    force_full_page_ocr=True,
    ocr_options=ocr_options
)

Example – manual language setting (e.g., Czech):

ocr_options = TesseractCliOcrOptions(lang=["ces", "eng"])

Important notes:

• For auto-detection, the osd.traineddata file must be installed, along with .traineddata files for any detected languages.
• The TESSDATA_PREFIX environment variable must be set before starting Python.
• The default docling-serve Docker image only includes eng and osd — a custom image is required for additional languages.