From 8076e30969e647df451a30e494d89c3ef4ffe99b Mon Sep 17 00:00:00 2001
From: Christoph Auer <cau@zurich.ibm.com>
Date: Tue, 25 Mar 2025 10:58:55 +0100
Subject: [PATCH] fix: Build lines from word level cells

Signed-off-by: Christoph Auer <cau@zurich.ibm.com>
---
 docling_parse/pdf_parser.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/docling_parse/pdf_parser.py b/docling_parse/pdf_parser.py
index 05524f6..f40401d 100644
--- a/docling_parse/pdf_parser.py
+++ b/docling_parse/pdf_parser.py
@@ -367,10 +367,13 @@ class PdfDocument:
         if len(segmented_page.textline_cells) > 0:
             return
 
+        self._create_word_cells(segmented_page, _loglevel)
+
         sanitizer = pdf_sanitizer(level=_loglevel)
 
         char_data = []
-        for item in segmented_page.char_cells:
+        # Note: We build the lines from the word level cells.
+        for item in segmented_page.word_cells:
             item_dict = item.model_dump(mode="json", by_alias=True, exclude_none=True)
 
             # TODO changing representation for the C++ parser, need to update on C++ code.