Spaces:
Sleeping
Sleeping
File size: 21,793 Bytes
162cb6f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 | """
structural_segmenter.py
=======================
Permissive segmenter for plays and novels. Handles smushed lines.
Enhancements over v1:
- Emotion tagging on dialogue blocks (lazily loaded EmotionAnalyzer)
- Better character cue parsing (handles "ROMEO.", "LADY MACBETH", "FIRST WITCH")
- French play/novel headings (ACTE, SCΓNE, CHAPITRE, PARTIE)
- Stage direction parsing improved (handles both () and [])
- Novel paragraph grouping: consecutive lines merged into paragraphs
- language parameter propagated through for multilingual emotion analysis
"""
from __future__ import annotations
import re
import uuid
from collections import namedtuple
from typing import Any, Dict, List, Optional, Tuple
import os
from services.hf_inference_service import HFInferenceService
# Initialize HF Inference Service
_hf_inference = HFInferenceService(os.getenv("HF_API_TOKEN"))
def _ai_is_heading(text: str) -> bool:
"""
Ask the cloud AI whether ``text`` is a chapter/section heading.
Only called for short (β€ 80 char), isolated text blocks.
"""
if os.getenv("USE_HF_INFERENCE") == "1" and _hf_inference.api_token:
try:
return _hf_inference.is_heading(text)
except Exception as e:
print(f"β οΈ AI heading check error: {e}")
pass
return False
# ββ Data types ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Span = namedtuple("Span", ["text", "size", "flags", "page"])
# ββ Helpers βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _uid() -> str:
return str(uuid.uuid4())[:8]
# ββ Heading regexes βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# English play headings
_ACT_RE = re.compile(r"\bACT\s+([IVX]+|\d+)\b", re.IGNORECASE)
_SCENE_RE = re.compile(r"\bSCENE\s+([IVX]+|\d+)\b", re.IGNORECASE)
# Abbreviated scene (e.g. Folger Shakespeare "SC. 2")
_SCENE_SC_RE = re.compile(r"\bSC\.\s*(\d+|[IVX]+)\b", re.IGNORECASE)
# French play headings
_ACTE_RE = re.compile(r"\bACTE\s+([IVX]+|\d+)\b", re.IGNORECASE)
_SCENE_FR_RE = re.compile(r"\bSC[ΓE]NE\s+([IVX]+|\d+)\b", re.IGNORECASE)
# Running-header detector: lines starting or ending with a page number,
# or containing book title artifacts.
_RUNNING_HEADER_RE = re.compile(
r"^\d{1,4}\s+\S|" # "9 MACBETH..."
r"^.*\s+\d{1,4}$|" # "...Before Breakfast 3"
r"^[A-Z\s]{5,}\s+\d{1,4}$" # "CHARLOTTE'S WEB 4"
)
# English novel headings
_CHAPTER_RE = re.compile(
r"\b(CHAPTER|PROLOGUE|EPILOGUE|PART|BOOK)\s+([\dIVX]+|[A-Z][a-z]+)?\b",
re.IGNORECASE,
)
# French novel headings
_CHAPITRE_RE = re.compile(r"\bCHAPITRE\s+([\dIVX]+|[A-Z][a-z]+)?\b", re.IGNORECASE)
_PARTIE_RE = re.compile(r"\bPARTIE\s+([\dIVX]+|[A-Z][a-z]+)?\b", re.IGNORECASE)
_LIVRE_RE = re.compile(r"\bLIVRE\s+([\dIVX]+|[A-Z][a-z]+)?\b", re.IGNORECASE)
# Character cue: ALL CAPS line (1β5 words), optional period/colon, <30 chars total
# Handles: "ROMEO", "LADY MACBETH", "FIRST WITCH.", "THE GHOST:"
_CUE_RE = re.compile(
r"^[A-Z][A-Z\s'\.\d]{0,28}[A-Z]\.?\:?\s*$"
)
# Stage directions
_STAGE_INLINE_RE = re.compile(r"^\[.*\]$|^\(.*\)$")
# Default neutral animation block
_NEUTRAL_ANIM = {
"expression": "neutral",
"eyebrows": "neutral",
"mouth": "closed",
"eyes": "normal",
"color_tint": "#f1f5f9",
}
# ββ Segmenter βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class StructuralSegmenter:
"""
Converts a flat list of PyMuPDF block dicts into a hierarchical structure:
- Plays: [Act β [Scene β [blocks]]]
- Novels: [Chapter β [Section β paragraphs]]
After segmentation, dialogue blocks in plays can be optionally enriched
with emotion data via the EmotionAnalyzer singleton.
"""
# ββ Public API βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def segment(
self,
all_blocks: List[Dict[str, Any]],
doc_type: str,
language: str = "en",
add_emotions: bool = True,
ai_headings: Optional[List[Dict[str, str]]] = None,
) -> List[Dict[str, Any]]:
"""
Segment blocks into hierarchical units.
Parameters
----------
all_blocks: List of PyMuPDF block dicts.
doc_type: "play" | "novel" | "generic"
language: "en" | "fr"
add_emotions: If True, run EmotionAnalyzer on dialogue blocks (plays only).
ai_headings: Optional list of metadata-inferred headings [{"title": "...", "text": "..."}].
"""
self.ai_headings = ai_headings or []
tokens = self._tokenise(all_blocks, doc_type)
if doc_type == "play":
units = self._build_play_hierarchy(tokens)
if add_emotions:
self._enrich_play_emotions(units, language)
return units
if doc_type == "poem":
return self._build_poem_hierarchy(tokens)
return self._build_novel_hierarchy(tokens)
# ββ Heading threshold βββββββββββββββββββββββββββββββββββββββββββββββββββ
@staticmethod
def _compute_heading_threshold(spans: List[Span]) -> float:
"""Compute a font-size threshold above which text is likely a heading."""
if not spans:
return 14.0
sizes = sorted(s.size for s in spans)
n = len(sizes)
median = sizes[n // 2] if n % 2 else (sizes[n // 2 - 1] + sizes[n // 2]) / 2
max_size = sizes[-1]
return median + (max_size - median) * 0.5
# ββ Tokenisation ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _tokenise(
self, blocks: List[Dict[str, Any]], doc_type: str
) -> List[Dict[str, Any]]:
tokens: List[Dict[str, Any]] = []
for b in blocks:
if b.get("type") != 0:
continue
for line_obj in b.get("lines", []):
txt = "".join(
s.get("text", "") for s in line_obj.get("spans", [])
).strip()
if not txt:
continue
level, matched = self._match_heading(txt, doc_type)
if matched:
tokens.append({
"type": "heading",
"level": level,
"title": txt,
"inferred": False,
})
else:
tokens.append({"type": "content", "text": txt})
return tokens
def _match_heading(self, text: str, doc_type: str) -> Tuple[str, bool]:
"""
Check for heading patterns.
Priority: AI-discovered headings -> Regex patterns.
"""
clean_text = self._normalize_heading(text)
# 1. AI Anchor Match (highest priority)
if hasattr(self, "ai_headings") and self.ai_headings:
for ah in self.ai_headings:
target = self._normalize_heading(ah.get("text", ah.get("title", "")))
if target and target in clean_text:
# Treat as chapter-level for novels/poems, act/scene follows regex logic below
return "chapter", True
t = text[:30]
# 2. Strong Regex Patterns (must check BEFORE running header as "CHAPTER 2" matches running header)
if doc_type == "play":
if _ACT_RE.search(t): return "act", True
if _ACTE_RE.search(t): return "act", True
if _SCENE_RE.search(t): return "scene", True
if _SCENE_SC_RE.search(t): return "scene", True
if _SCENE_FR_RE.search(t): return "scene", True
else:
if _CHAPTER_RE.search(t): return "chapter", True
if _CHAPITRE_RE.search(t): return "chapter", True
if _PARTIE_RE.search(t): return "chapter", True
if _LIVRE_RE.search(t): return "chapter", True
# 3. Skip running headers (page numbers, book titles at edges)
if _RUNNING_HEADER_RE.match(text.strip()):
return "none", False
# 4. Mistral-7B fallback: ask the model for short, isolated text
# that looks like it could be an unnamed heading (e.g. "The Storm",
# "I.", roman numerals, numbered titles without the word CHAPTER).
if len(text.strip()) <= 80 and _ai_is_heading(text.strip()):
return "chapter", True
return "none", False
@staticmethod
def _normalize_heading(text: str) -> str:
"""Lowercases and strips common punctuation/whitespace for comparison."""
return re.sub(r'[^a-zA-Z0-9]', '', text.lower())
# ββ Play hierarchy βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _build_play_hierarchy(
self, tokens: List[Dict[str, Any]]
) -> List[Dict[str, Any]]:
acts: List[Dict[str, Any]] = []
cur_act: Optional[Dict] = None
cur_scene: Optional[Dict] = None
content_buffer: List[str] = []
def flush():
if cur_scene is not None and content_buffer:
cur_scene["blocks"].extend(
self._parse_play_content(content_buffer)
)
content_buffer.clear()
for tok in tokens:
if tok["type"] == "heading":
if tok["level"] == "act":
flush()
if cur_act and cur_scene:
cur_act["children"].append(cur_scene)
if cur_act:
acts.append(cur_act)
cur_act = {"id": _uid(), "title": tok["title"], "children": []}
cur_scene = None
elif tok["level"] == "scene":
flush()
if cur_act and cur_scene:
cur_act["children"].append(cur_scene)
if cur_act is None:
cur_act = {"id": _uid(), "title": "ACT I", "children": [], "inferred": True}
cur_scene = {"id": _uid(), "title": tok["title"], "blocks": []}
else: # content token
if cur_act is None:
continue
if cur_scene is None:
cur_scene = {
"id": _uid(), "title": "Scene 1", "blocks": [],
"inferred": True,
}
content_buffer.append(tok["text"])
# Flush final scene/act
flush()
if cur_act and cur_scene:
cur_act["children"].append(cur_scene)
if cur_act:
acts.append(cur_act)
return acts if acts else self._play_fallback(tokens)
def _parse_play_content(self, lines: List[str]) -> List[Dict[str, Any]]:
"""
Convert flat text lines into structured block dicts.
Block types:
"dialogue" β character speech
"stage_direction" β [Enter Juliet] or (aside)
"narrative" β other text
"""
blocks: List[Dict[str, Any]] = []
for line in lines:
stripped = line.strip()
if not stripped:
continue
# Strip leading page numbers smushed with text (e.g. "11She's" -> "She's")
stripped = re.sub(r'^\d{1,3}([A-Z])', r'\1', stripped)
# Skip running headers
if _RUNNING_HEADER_RE.match(stripped) or _RUNNING_HEADER_RE.search(stripped):
if len(stripped) < 50: # Headers are usually short
continue
# Stage direction: entire line wrapped in () or []
if _STAGE_INLINE_RE.match(stripped):
blocks.append({
"type": "stage_direction",
"character": None,
"content": stripped.strip("[]()").strip(),
})
continue
# Character cue (ALL CAPS, β€6 words, short line)
word_count = len(stripped.split())
if _CUE_RE.match(stripped) and word_count <= 6:
cue = stripped.rstrip(":. \t")
blocks.append({
"type": "dialogue",
"character": cue,
"content": "",
"emotion": "neutral",
"intensity": 0.5,
"anim": dict(_NEUTRAL_ANIM),
})
continue
# Append to last dialogue block if mid-speech
if blocks and blocks[-1]["type"] == "dialogue":
# Skip bare page numbers (e.g. "5", "12") slipping into speech
if re.match(r"^\d{1,4}$", stripped):
continue
sep = " " if blocks[-1]["content"] else ""
blocks[-1]["content"] += sep + stripped
continue
# Narrative line
blocks.append({"type": "narrative", "content": stripped})
return blocks
def _play_fallback(self, tokens: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
"""Single-unit fallback when no act/scene headings found."""
text_lines = [t["text"] for t in tokens if t["type"] == "content"]
return [{
"id": _uid(),
"title": "The Play",
"children": [{
"id": _uid(),
"title": "Content",
"blocks": self._parse_play_content(text_lines),
}],
}]
# ββ Novel hierarchy ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _build_novel_hierarchy(
self, tokens: List[Dict[str, Any]]
) -> List[Dict[str, Any]]:
chapters: List[Dict[str, Any]] = []
MAX_CHUNK_WORDS = 250
def new_chapter(title: str) -> Dict[str, Any]:
return {
"id": _uid(),
"title": title,
"children": [], # We will add chunks here
"content": "",
}
def new_chunk(title: str = "") -> Dict[str, Any]:
return {
"id": _uid(),
"title": title,
"paragraphs": [],
"content": "",
"word_count": 0
}
cur_chapter = new_chapter("Beginning")
cur_chunk = new_chunk("Introduction")
for tok in tokens:
if tok["type"] == "heading" and tok["level"] == "chapter":
# Flush current chunk and chapter
if cur_chunk["paragraphs"]:
cur_chapter["children"].append(cur_chunk)
if cur_chapter["children"]:
self._finalize_chapter(cur_chapter)
chapters.append(cur_chapter)
cur_chapter = new_chapter(tok["title"])
cur_chunk = new_chunk()
elif tok["type"] == "content":
text = tok["text"]
words = len(text.split())
# If adding this would explode the chunk, flush first
if cur_chunk["word_count"] > 0 and (cur_chunk["word_count"] + words) > MAX_CHUNK_WORDS:
cur_chapter["children"].append(cur_chunk)
cur_chunk = new_chunk()
cur_chunk["paragraphs"].append(text)
cur_chunk["word_count"] += words
# Final flush
if cur_chunk["paragraphs"]:
cur_chapter["children"].append(cur_chunk)
if cur_chapter["children"]:
self._finalize_chapter(cur_chapter)
chapters.append(cur_chapter)
return chapters
@staticmethod
def _finalize_chapter(chapter: Dict[str, Any]):
"""Finalize all chunks in the chapter and set overall content."""
all_content = []
for i, chunk in enumerate(chapter["children"]):
chunk["content"] = "\n\n".join(chunk.get("paragraphs", []))
# If no specific title, give it a sequence title
if not chunk["title"]:
chunk["title"] = f"Part {i+1}"
all_content.append(chunk["content"])
chapter["content"] = "\n\n".join(all_content)
# ββ Poem hierarchy βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _build_poem_hierarchy(
self, tokens: List[Dict[str, Any]]
) -> List[Dict[str, Any]]:
"""
Build stanza-based hierarchy for poems.
Groups consecutive content lines into stanzas, separated by blank gaps
or heading tokens.
"""
poems: List[Dict[str, Any]] = []
current_poem: Optional[Dict] = None
current_stanza: List[str] = []
stanza_num = 0
def flush_stanza():
nonlocal stanza_num
if current_poem is not None and current_stanza:
stanza_num += 1
current_poem["children"].append({
"id": _uid(),
"title": f"Stanza {stanza_num}",
"blocks": [
{"type": "verse_line", "content": line}
for line in current_stanza
],
"content": "\n".join(current_stanza),
})
current_stanza.clear()
for tok in tokens:
if tok["type"] == "heading":
flush_stanza()
if current_poem and current_poem["children"]:
poems.append(current_poem)
current_poem = {"id": _uid(), "title": tok["title"], "children": []}
stanza_num = 0
else:
text = tok.get("text", "").strip()
if not text:
# Empty line = stanza break
flush_stanza()
continue
if current_poem is None:
current_poem = {"id": _uid(), "title": "The Poem", "children": []}
current_stanza.append(text)
# Flush remaining
flush_stanza()
if current_poem and current_poem["children"]:
poems.append(current_poem)
if not poems:
# Fallback: treat all content as one poem
all_lines = [t["text"] for t in tokens if t["type"] == "content"]
stanzas = []
current: List[str] = []
for line in all_lines:
if not line.strip():
if current:
stanzas.append(current)
current = []
else:
current.append(line)
if current:
stanzas.append(current)
children = [
{
"id": _uid(),
"title": f"Stanza {i + 1}",
"blocks": [{"type": "verse_line", "content": l} for l in st],
"content": "\n".join(st),
}
for i, st in enumerate(stanzas)
] if stanzas else [{
"id": _uid(),
"title": "Content",
"blocks": [{"type": "verse_line", "content": l} for l in all_lines],
"content": "\n".join(all_lines),
}]
poems = [{
"id": _uid(),
"title": "The Poem",
"children": children,
}]
return poems
# ββ Emotion enrichment ββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _enrich_play_emotions(
self, acts: List[Dict[str, Any]], language: str = "en"
):
"""
Walk Act β Scene β Block tree and add emotion data to all dialogue blocks.
"""
try:
from .emotion_analyzer import get_emotion_analyzer
ea = get_emotion_analyzer()
except Exception as e:
print(f"β οΈ StructuralSegmenter: emotion enrichment skipped ({e})")
return
for act in acts:
for scene in act.get("children", []):
blocks = scene.get("blocks", [])
try:
ea.enrich_dialogue_blocks(blocks, language=language)
except Exception as exc:
print(
f"β οΈ Emotion error in scene "
f"'{scene.get('title', '?')}': {exc}"
)
|