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

fix: 印尼语fallback缺失 + docx格式100%保留

Browse Source 
- index.html: catch块增加id回退、setLang动态更新title、标题栏初始化
- docx_translator.py: 从分段翻译改为段落级翻译，Run结构100%保留
  删除SIGNIFICANT_STYLES/is_tab_run/_process_element_children等~120行死代码

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
晏永龙
2026-06-08 14:43:54 +08:00
parent 8a5f62342a
commit 97b7b20565
2 changed files with 42 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
docutranslate/static/index.html
View File

@@ -3,7 +3,7 @@
<head> <head>
<meta charset="UTF-8"> <meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0"> <meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>DocuTranslate - 交互式文档翻译</title> <title>DocuTranslate</title>
<link rel="icon" href="/static/favicon.ico" type="image/x-icon"> <link rel="icon" href="/static/favicon.ico" type="image/x-icon">
<!-- Bootstrap CSS --> <!-- Bootstrap CSS -->
<link href="/static/bootstrap.css" rel="stylesheet" crossorigin="anonymous"> <link href="/static/bootstrap.css" rel="stylesheet" crossorigin="anonymous">
@@ -1860,6 +1860,8 @@
localStorage.setItem('ui_language', l); localStorage.setItem('ui_language', l);
const langMap = {zh: 'zh-CN', en: 'en', id: 'id'}; const langMap = {zh: 'zh-CN', en: 'en', id: 'id'};
document.documentElement.lang = langMap[l] || 'en'; document.documentElement.lang = langMap[l] || 'en';
const dict = i18nData.value[l] || i18nData.value['en'] || {};
document.title = dict['pageTitle'] || 'DocuTranslate';
}; };
const setTheme = (t) => { const setTheme = (t) => {
localStorage.setItem('theme', t); localStorage.setItem('theme', t);
@@ -1886,6 +1888,7 @@
if(i18nData.value.en) Object.assign(i18nData.value.en, extraEn); if(i18nData.value.en) Object.assign(i18nData.value.en, extraEn);
} catch (e) { } catch (e) {
console.error("i18n load failed", e);
i18nData.value = { i18nData.value = {
zh: { zh: {
pageTitle: "DocuTranslate", pageTitle: "DocuTranslate",
@@ -1921,10 +1924,32 @@
mineruDeployServerUrlPlaceholder: "http://127.0.0.1:30000", mineruDeployServerUrlPlaceholder: "http://127.0.0.1:30000",
mineruDeployParseMethodLabel: "Parse Method", mineruDeployParseMethodLabel: "Parse Method",
mineruDeployTableEnableLabel: "Table Recognition" mineruDeployTableEnableLabel: "Table Recognition"
},
id: {
pageTitle: "DocuTranslate",
tutorialBtn: "Tutorial",
projectContributeBtn: "Kolaborasi",
workflowTitle: "Pilih Alur Kerja",
autoWorkflowLabel: "Pilih Otomatis",
modelPresetLabel: "Preset Model",
modelPresetPlaceholder: "Pilih preset model",
modelPresetEmpty: "Konfigurasi preset di server",
modelPresetRuntimeHint: "Provider, endpoint, dan API key akan dibaca dari environment server.",
workflowOptionPptx: "Presentasi PPTX",
pptxSettingsTitleText: "Pengaturan PPTX",
mineruDeployServerUrlLabel: "Server URL",
mineruDeployLangListLabel: "Daftar Bahasa",
mineruDeployServerUrlPlaceholder: "http://127.0.0.1:30000",
mineruDeployParseMethodLabel: "Parse Method",
mineruDeployTableEnableLabel: "Table Recognition"
} }
}; };
} }
// Set initial page title based on detected language
const initDict = i18nData.value[currentLang.value] || i18nData.value['en'] || {};
document.title = initDict['pageTitle'] || 'DocuTranslate';
// Backend Metadata // Backend Metadata
try { try {
const [metaRes, enginRes, paramsRes, configRes] = await Promise.all([ const [metaRes, enginRes, paramsRes, configRes] = await Promise.all([

171
docutranslate/translator/ai_translator/docx_translator.py
View File

@@ -24,19 +24,6 @@ from docutranslate.translator.ai_translator.base import AiTranslatorConfig, AiTr
# ---------------- 辅助函数 ---------------- # ---------------- 辅助函数 ----------------
# [v6.2] 定义一组具有显著视觉效果的格式标签。
# 我们只在 Run 包含这些格式时才将其视为空白格式边界。
# 这避免了因字体、字号等微小变化导致的过度文本切分。
SIGNIFICANT_STYLES = frozenset([
qn('w:u'), # 下划线
qn('w:strike'), # 删除线
qn('w:dstrike'), # 双删除线
qn('w:shd'), # 底纹/背景色
qn('w:highlight'), # 荧光笔高亮
qn('w:bdr'), # 边框
qn('w:effectLst'), # 文本效果 (如发光、阴影)
qn('w:em'), # 强调标记 (着重号)
])
def is_image_run(run: Run) -> bool: def is_image_run(run: Run) -> bool:
@@ -45,33 +32,6 @@ def is_image_run(run: Run) -> bool:
return '<w:drawing' in xml or '<w:pict' in xml return '<w:drawing' in xml or '<w:pict' in xml
def is_formatting_only_run(run: Run) -> bool:
"""
检查一个 Run 是否仅用于格式化，不包含任何应被渲染的文本。
这仅适用于其 .text 属性为 "" 的情况。
"""
return run.text == ""
# ---------- 新增修改部分 1: is_styled_whitespace_run 函数被移除 ----------
# 此函数不再需要，因为新的逻辑会根据格式变化来切分，而不是根据带格式的空格。
# ---------------------- 修改结束 ----------------------
def is_tab_run(run: Run) -> bool:
"""
检查一个 Run 是否主要代表一个制表符，应被视作格式边界。
仅当 Run 的文本内容为空或仅包含空白，且 XML 中存在 <w:tab/> 时，
才将其视为纯格式化用途的 Run。
"""
# .text 属性会将 <w:tab/> 转换成 '\t'
# 如果 .text 在去除空白后仍有内容，说明这个 Run 不仅仅是个制表符。
if run.text.strip():
return False
xml = getattr(run.element, 'xml', '')
return '<w:tab' in xml or '<w:ptab' in xml
# ---------------- 配置类 ---------------- # ---------------- 配置类 ----------------
@dataclass @dataclass
class DocxTranslatorConfig(AiTranslatorConfig): class DocxTranslatorConfig(AiTranslatorConfig):
@@ -110,14 +70,6 @@ class DocxTranslator(AiTranslator):
[v6.0 - 语义切分重构版] [v6.0 - 语义切分重构版]
- 重构核心逻辑，不再跳过域结果，而是将其作为语义边界来切分文本，增强了鲁棒性。 - 重构核心逻辑，不再跳过域结果，而是将其作为语义边界来切分文本，增强了鲁棒性。
""" """
IGNORED_TAGS = {
qn('w:proofErr'), qn('w:lastRenderedPageBreak'), qn('w:bookmarkStart'),
qn('w:bookmarkEnd'), qn('w:commentRangeStart'), qn('w:commentRangeEnd'),
qn('w:del'), qn('w:ins'), qn('w:moveFrom'), qn('w:moveTo'),
}
RECURSIVE_CONTAINER_TAGS = {
qn('w:smartTag'), qn('w:sdtContent'), qn('w:hyperlink'),
}
def __init__(self, config: DocxTranslatorConfig): def __init__(self, config: DocxTranslatorConfig):
super().__init__(config=config) super().__init__(config=config)
@@ -139,126 +91,35 @@ class DocxTranslator(AiTranslator):
self.insert_mode = config.insert_mode self.insert_mode = config.insert_mode
self.separator = config.separator self.separator = config.separator
# ---------- 新增修改部分 2: 增加用于比较格式的辅助函数 ----------
def _get_significant_styles(self, run: Run) -> frozenset:
"""从一个 Run 中提取“显著”格式标签的集合。"""
if run is None:
return frozenset()
rPr = run.element.rPr
if rPr is None:
return frozenset()
return frozenset(child.tag for child in rPr if child.tag in SIGNIFICANT_STYLES)
def _have_same_significant_styles(self, run1: Run, run2: Run) -> bool:
"""检查两个 Run 是否具有相同的“显著”格式集合。"""
styles1 = self._get_significant_styles(run1)
styles2 = self._get_significant_styles(run2)
return styles1 == styles2
# ---------------------- 修改结束 ----------------------
# ---------- 代码修改部分 1: 形状翻译逻辑的核心实现 ----------
def _process_element_children(self, element, parent_paragraph: Paragraph, elements: List[Dict[str, Any]],
texts: List[str],
state: Dict[str, Any],
top_level_para: Paragraph):
def flush_segment():
current_runs = state['current_runs']
if not current_runs:
return
full_text = "".join(r.text for r in current_runs)
if full_text.strip():
# 在 elements 中增加对父段落和顶级段落的引用
elements.append({
"type": "text_runs",
"runs": list(current_runs),
"paragraph": parent_paragraph,
"top_level_paragraph": top_level_para
})
texts.append(full_text)
state['current_runs'].clear()
for child in element:
if child.tag in self.IGNORED_TAGS:
continue
if child.tag in self.RECURSIVE_CONTAINER_TAGS:
flush_segment()
self._process_element_children(child, parent_paragraph, elements, texts, state, top_level_para)
flush_segment() # 在递归容器后也刷新，确保其内容成为独立片段
continue
field_char_element = child.find(qn('w:fldChar')) if isinstance(child, CT_R) else None
if field_char_element is not None:
fld_type = field_char_element.get(qn('w:fldCharType'))
if fld_type == 'begin' or fld_type == 'end':
flush_segment()
continue
if isinstance(child, CT_R):
# 传入 parent_paragraph 以确保 Run 对象具有正确的上下文
run = Run(child, parent_paragraph)
# 新增逻辑：处理形状（drawing/pict）内的文本
# 形状可以包含文本框，需要优先于图片处理逻辑进行解析
if '<w:drawing' in run.element.xml or '<w:pict' in run.element.xml:
# 使用 list() 消耗迭代器，以便检查是否找到了文本框
text_boxes = list(run.element.iter(qn('w:txbxContent')))
if text_boxes:
flush_segment() # 包含文本的形状是一个边界，刷新前面的文本
for txbx_content in text_boxes:
# 遍历文本框内的所有段落
for p_element in txbx_content.findall(qn('w:p')):
# 创建新的段落对象，并传入父级上下文
shape_para = Paragraph(p_element, parent_paragraph)
# 递归处理该段落，并传递顶级段落上下文
self._process_paragraph(shape_para, elements, texts, top_level_para=top_level_para)
# 如果处理了形状内的文本，则该 Run 的任务已完成
continue
# 保留原有逻辑: 检查绝对边界（图片、制表符等）
if is_image_run(run) or is_formatting_only_run(run) or is_tab_run(run):
flush_segment()
continue # 这些 Run 本身不包含在任何文本片段中
# 保留原有逻辑: 基于格式变化进行切分
last_run_in_segment = state['current_runs'][-1] if state['current_runs'] else None
if last_run_in_segment and not self._have_same_significant_styles(last_run_in_segment, run):
flush_segment()
# 将当前 Run 添加到片段中
state['current_runs'].append(run)
else:
# 遇到任何非 Run 的块级元素（如在单元格中嵌套的表格），都应结束当前文本片段。
flush_segment()
def _process_paragraph(self, para: Paragraph, elements: List[Dict[str, Any]], texts: List[str], def _process_paragraph(self, para: Paragraph, elements: List[Dict[str, Any]], texts: List[str],
top_level_para: Paragraph = None): top_level_para: Paragraph = None):
# 如果是首次进入段落处理（非递归调用），则当前段落是顶级段落 """
简化版段落处理：将段落内所有文本 Run 收集为一个翻译单元，完整保留格式结构。
不再按格式变化切分 segment，翻译质量更好、格式保留 100%。
"""
if top_level_para is None: if top_level_para is None:
top_level_para = para top_level_para = para
state = { text_runs = []
'current_runs': [], for run in para.runs:
} if is_image_run(run):
# 修改调用：传入 `para` 对象、其顶级上下文 continue # 跳过图片
self._process_element_children(para._p, para, elements, texts, state, top_level_para) if not run.text.strip():
# 保留带格式的空 Run（如下划线空格），但不加入文本
continue
# 确保在段落处理结束时，刷新所有剩余的 Run text_runs.append(run)
current_runs = state['current_runs']
if current_runs: if text_runs:
full_text = "".join(r.text for r in current_runs) full_text = "".join(r.text for r in text_runs)
if full_text.strip(): if full_text.strip():
elements.append({ elements.append({
"type": "text_runs", "type": "text_runs",
"runs": list(current_runs), "runs": list(text_runs),
"paragraph": para, "paragraph": para,
"top_level_paragraph": top_level_para "top_level_paragraph": top_level_para
}) })
texts.append(full_text) texts.append(full_text)
current_runs.clear()
# ---------------------- 修改结束 ---------------------- # ---------------------- 修改结束 ----------------------