Leon/docutranslate
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

优化docx、提示词

Browse Source
This commit is contained in:
xunbu
2025-10-13 20:51:53 +08:00
parent 11c8be3c49
commit 9676fdaf99
3 changed files with 100 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

137
docutranslate/translator/ai_translator/docx_translator.py
View File

@@ -55,9 +55,29 @@ class DocxTranslator(AiTranslator):
"""
用于翻译 .docx 文件的高级翻译器，能够高精度保留样式、处理超链接、
域代码（如图注），并支持翻译脚注、尾注等。
[v3.1 - 采纳新规则：将带格式的空文本视为分割点]
[v3.6 - 引入递归解析以处理嵌套内容标签，如 smartTag]
"""
# 包含所有应被解析器完全忽略的、不影响文本内容的元数据标签
IGNORED_TAGS = {
qn('w:proofErr'), # 拼写和语法错误标记
qn('w:lastRenderedPageBreak'), # 上次渲染的分页符位置
qn('w:bookmarkStart'), # 书签开始
qn('w:bookmarkEnd'), # 书签结束
qn('w:commentRangeStart'), # 批注范围开始
qn('w:commentRangeEnd'), # 批注范围结束
qn('w:del'), # 修订：删除
qn('w:ins'), # 修订：插入
qn('w:moveFrom'), # 修订：移动源
qn('w:moveTo'), # 修订：移动目标
}
# 包含应递归处理其内部内容的容器标签
RECURSIVE_CONTAINER_TAGS = {
qn('w:smartTag'), # 智能标记 (包含文本)
qn('w:sdtContent'), # 结构化文档标签内容 (包含文本)
}
def __init__(self, config: DocxTranslatorConfig):
super().__init__(config=config)
self.chunk_size = config.chunk_size
@@ -104,16 +124,13 @@ class DocxTranslator(AiTranslator):
pass
return False
def _process_paragraph(self, para: Paragraph, elements: List[Dict[str, Any]], texts: List[str]):
def _process_element_children(self, element, elements: List[Dict[str, Any]], texts: List[str],
state: Dict[str, Any]):
"""
使用状态机处理段落，将连续的文本Run合并为一个翻译单元，
同时将图片、超链接、带格式的空文本等视为分割点。
[新函数] 递归处理任何给定XML元素的子节点。
'state' 字典用于跨递归调用传递状态，如 current_runs 和 is_inside_field。
"""
if not para.text.strip():
return
current_runs = []
is_inside_field = False
current_runs = state['current_runs']
def flush_segment():
nonlocal current_runs
@@ -124,43 +141,73 @@ class DocxTranslator(AiTranslator):
elements.append({"type": "text_runs", "runs": current_runs})
texts.append(full_text)
current_runs = []
state['current_runs'] = current_runs
for child in para._p:
if self._is_seq_field(child):
flush_segment()
for child in element:
if child.tag in self.IGNORED_TAGS:
continue
if child.tag == qn('w:fldChar'):
fld_type = child.get(qn('w:fldCharType'))
if fld_type == 'begin':
flush_segment()
is_inside_field = True
elif fld_type == 'end':
is_inside_field = False
if child.tag in self.RECURSIVE_CONTAINER_TAGS:
self._process_element_children(child, elements, texts, state)
continue
if is_inside_field:
continue
if isinstance(child, CT_R):
run = Run(child, para)
# 如果是图片、制表符，或者【带格式的空文本】，则视为分割点
if is_image_run(run) or run.element.find(qn('w:tab')) is not None or is_formatting_only_run(run):
flush_segment()
# 这个 run 本身被保留，不参与翻译
else:
# 否则，它是一个普通的文本 Run，收集起来
current_runs.append(run)
elif child.tag == qn('w:hyperlink'):
if child.tag == qn('w:hyperlink'):
flush_segment()
hyperlink_text = self._extract_hyperlink_text(child)
if hyperlink_text.strip():
elements.append({"type": "hyperlink", "element": child})
texts.append(hyperlink_text)
continue
field_char_element = None
if child.tag == qn('w:fldChar'):
field_char_element = child
elif isinstance(child, CT_R):
field_char_element = child.find(qn('w:fldChar'))
if field_char_element is not None:
flush_segment()
fld_type = field_char_element.get(qn('w:fldCharType'))
if fld_type == 'begin':
state['is_inside_field'] = True
elif fld_type == 'end':
state['is_inside_field'] = False
continue
if state.get('is_inside_field', False):
continue
if self._is_seq_field(child):
flush_segment()
continue
if isinstance(child, CT_R):
run = Run(child, None)
if is_image_run(run) or run.element.find(qn('w:tab')) is not None or is_formatting_only_run(run):
flush_segment()
else:
current_runs.append(run)
else:
flush_segment()
flush_segment()
state['current_runs'] = current_runs
def _process_paragraph(self, para: Paragraph, elements: List[Dict[str, Any]], texts: List[str]):
"""
[重构] 作为递归处理器的入口点，初始化状态并调用递归函数。
"""
if not para.text.strip():
return
state = {'current_runs': [], 'is_inside_field': False}
self._process_element_children(para._p, elements, texts, state)
current_runs = state['current_runs']
if current_runs:
full_text = "".join(r.text for r in current_runs)
if full_text.strip():
elements.append({"type": "text_runs", "runs": current_runs})
texts.append(full_text)
def _process_container(self, container, elements: List[Dict[str, Any]], texts: List[str]):
"""递归处理包含段落和表格的容器（如文档、单元格、页眉）。"""
@@ -204,14 +251,23 @@ class DocxTranslator(AiTranslator):
return doc, elements, texts
def _apply_translation(self, element_info: Dict[str, Any], final_text: str):
"""将翻译后的文本写回对应的 OXML 元素。对于多Run的文本段，写入第一个Run并清空其余。"""
"""
将翻译后的文本写回对应的 OXML 元素。
对于多Run的文本段，写入第一个Run并【删除】其余，以避免产生方框占位符。
"""
el_type = element_info["type"]
if el_type == "text_runs":
runs = element_info["runs"]
if runs:
runs[0].text = final_text
for run in runs[1:]:
run.text = ""
if not runs:
return
runs[0].text = final_text
for run in runs[1:]:
p_element = run.element.getparent()
if p_element is not None:
p_element.remove(run.element)
elif el_type == "hyperlink":
hyperlink = element_info["element"]
r_elements = hyperlink.findall(f'.//{qn("w:r")}')
@@ -219,10 +275,12 @@ class DocxTranslator(AiTranslator):
first_r = r_elements[0]
for t in first_r.findall(f'.//{qn("w:t")}'):
first_r.remove(t)
new_t = OxmlElement('w:t')
new_t.text = final_text
new_t.set(qn('xml:space'), 'preserve')
first_r.append(new_t)
for other_r in r_elements[1:]:
if (parent := other_r.getparent()) is not None:
parent.remove(other_r)
@@ -230,7 +288,8 @@ class DocxTranslator(AiTranslator):
def _after_translate(self, doc: DocumentObject, elements: List[Dict[str, Any]], translated: List[str],
originals: List[str]) -> bytes:
if len(elements) != len(translated):
self.logger.error(f"Translation count mismatch! Originals: {len(originals)}, Translated: {len(translated)}. Processing common part only.")
self.logger.error(
f"Translation count mismatch! Originals: {len(originals)}, Translated: {len(translated)}. Processing common part only.")
min_len = min(len(elements), len(translated), len(originals))
elements, translated, originals = elements[:min_len], translated[:min_len], originals[:min_len]