Preserve newlines in Table and TableChunk elements during PDF partitioning

CHANGELOG.md

@@ -1,8 +1,11 @@
-## 0.18.33-dev0
+## 0.18.33-dev1
 
 ### Enhancements
 - **Add `group_elements_by_parent_id` utility function**: Groups elements by their `parent_id` metadata field for easier document hierarchy traversal (fixes #1489)
 
+### Fixes
+- **Preserve newlines in Table/TableChunk elements during PDF partitioning**: Skip whitespace normalization for Table and TableChunk elements so newlines that carry structural meaning (row separation) are preserved (fixes #3983)
+
 ## 0.18.32
 
 ### Enhancements

unstructured/__version__.py

@@ -1 +1 @@
-__version__ = "0.18.33-dev0"  # pragma: no cover
+__version__ = "0.18.33-dev1"  # pragma: no cover

unstructured/partition/pdf.py

@@ -34,6 +34,8 @@
     Link,
     ListItem,
     PageBreak,
+    Table,
+    TableChunk,
     Text,
     Title,
 )
@@ -823,11 +825,16 @@ def _partition_pdf_or_image_local(
             out_elements.append(cast(Element, el))
         # NOTE(crag): this is probably always a Text object, but check for the sake of typing
         elif isinstance(el, Text):
-            el.text = re.sub(
-                RE_MULTISPACE_INCLUDING_NEWLINES,
-                " ",
-                el.text or "",
-            ).strip()
+            if isinstance(el, (Table, TableChunk)):
+                # For Table/TableChunk, preserve newlines (they carry structural meaning)
+                # but still collapse multiple horizontal whitespace (spaces, tabs) to single space
+                el.text = re.sub(r"[^\S\n]+", " ", el.text or "").strip()
+            else:
+                el.text = re.sub(
+                    RE_MULTISPACE_INCLUDING_NEWLINES,
+                    " ",
+                    el.text or "",
+                ).strip()
             if el.text or isinstance(el, PageBreak):
                 out_elements.append(cast(Element, el))