fix(pdf): allow configuring rotated text direction

README.md

@@ -110,6 +110,15 @@ At the moment, the following optional dependencies are available:
 * `[audio-transcription]` Installs dependencies for audio transcription of wav and mp3 files
 * `[youtube-transcription]` Installs dependencies for fetching YouTube video transcription
 
+For PDFs containing bottom-to-top rotated text, pass the direction expected by
+pdfplumber when converting the document:
+
+```python
+from markitdown import MarkItDown
+
+result = MarkItDown().convert("rotated.pdf", pdf_char_dir_rotated="btt")
+```
+
 ### Plugins
 
 MarkItDown also supports 3rd-party plugins. Plugins are disabled by default. To list installed plugins:

packages/markitdown/src/markitdown/converters/_pdf_converter.py

@@ -117,7 +117,9 @@ def fmt_row(row: list[str]) -> str:
     return "\n".join(md)
 
 
-def _extract_form_content_from_words(page: Any) -> str | None:
+def _extract_form_content_from_words(
+    page: Any, *, char_dir_rotated: str | None = None
+) -> str | None:
     """
     Extract form-style content from a PDF page by analyzing word positions.
     This handles borderless forms/tables where words are aligned in columns.
@@ -129,7 +131,14 @@ def _extract_form_content_from_words(page: Any) -> str | None:
     Returns None if the page doesn't appear to be a form-style document,
     indicating that pdfminer should be used instead for better text spacing.
     """
-    words = page.extract_words(keep_blank_chars=True, x_tolerance=3, y_tolerance=3)
+    extract_words_kwargs: dict[str, Any] = {
+        "keep_blank_chars": True,
+        "x_tolerance": 3,
+        "y_tolerance": 3,
+    }
+    if char_dir_rotated is not None:
+        extract_words_kwargs["char_dir_rotated"] = char_dir_rotated
+    words = page.extract_words(**extract_words_kwargs)
     if not words:
         return None
 
@@ -395,15 +404,24 @@ def extract_cells(info: dict) -> list[str]:
     return "\n".join(result_lines)
 
 
-def _extract_tables_from_words(page: Any) -> list[list[list[str]]]:
+def _extract_tables_from_words(
+    page: Any, *, char_dir_rotated: str | None = None
+) -> list[list[list[str]]]:
     """
     Extract tables from a PDF page by analyzing word positions.
     This handles borderless tables where words are aligned in columns.
 
     This function is designed for structured tabular data (like invoices),
     not for multi-column text layouts in scientific documents.
     """
-    words = page.extract_words(keep_blank_chars=True, x_tolerance=3, y_tolerance=3)
+    extract_words_kwargs: dict[str, Any] = {
+        "keep_blank_chars": True,
+        "x_tolerance": 3,
+        "y_tolerance": 3,
+    }
+    if char_dir_rotated is not None:
+        extract_words_kwargs["char_dir_rotated"] = char_dir_rotated
+    words = page.extract_words(**extract_words_kwargs)
     if not words:
         return []
 
@@ -536,6 +554,11 @@ def convert(
 
         assert isinstance(file_stream, io.IOBase)
 
+        char_dir_rotated = kwargs.get("pdf_char_dir_rotated")
+        extract_text_kwargs: dict[str, Any] = {}
+        if char_dir_rotated is not None:
+            extract_text_kwargs["char_dir_rotated"] = char_dir_rotated
+
         # Read file stream into BytesIO for compatibility with pdfplumber
         pdf_bytes = io.BytesIO(file_stream.read())
 
@@ -551,15 +574,17 @@ def convert(
 
             with pdfplumber.open(pdf_bytes) as pdf:
                 for page_idx, page in enumerate(pdf.pages):
-                    page_content = _extract_form_content_from_words(page)
+                    page_content = _extract_form_content_from_words(
+                        page, char_dir_rotated=char_dir_rotated
+                    )
 
                     if page_content is not None:
                         form_page_count += 1
                         if page_content.strip():
                             markdown_chunks.append(page_content)
                     else:
                         plain_page_indices.append(page_idx)
-                        text = page.extract_text()
+                        text = page.extract_text(**extract_text_kwargs)
                         if text and text.strip():
                             markdown_chunks.append(text.strip())
 

packages/markitdown/tests/test_pdf_tables.py

@@ -1196,3 +1196,15 @@ def test_borderless_table_data_integrity(self, markitdown):
         table_text = str(second_table)
         assert "Electronics" in table_text, "Second table should contain Electronics"
         assert "Hardware" in table_text, "Second table should contain Hardware"
+
+    def test_rotated_text_direction_can_be_configured(self, markitdown):
+        """Test that bottom-to-top rotated text can use pdfplumber's direction option."""
+        pdf_path = os.path.join(TEST_FILES_DIR, "rotated_table.pdf")
+
+        default_result = markitdown.convert(pdf_path)
+        assert "noitalupoP" in default_result.text_content
+
+        result = markitdown.convert(pdf_path, pdf_char_dir_rotated="btt")
+        assert "Projected" in result.text_content
+        assert "Population" in result.text_content
+        assert "noitalupoP" not in result.text_content