Spaces:
Running
Running
File size: 13,468 Bytes
be1ce5d | 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 | #!/usr/bin/env python3
"""
clean_corpora.py — Parse and normalize the four ghost corpora into entry-level JSONL.
Sources (all US public domain):
- Samuel Pepys, Diary (1660-1669). PG #4200, Wheatley ed. Footnotes stripped.
- Vincent van Gogh, Letters of a Post-Impressionist (Ludovici trans., 1912/1913). PG #40393.
- Katherine Mansfield, Journal (Knopf 1927/1928 printing). Archive.org OCR; editorial
brackets by J.M. Murry stripped (both for voice purity and conservatism).
- Mary MacLane, The Story of Mary MacLane (1902) PG #43696, and
I, Mary MacLane (1917) PG #43556.
Output: data/clean/<persona>.jsonl with one entry per line:
{"persona": str, "date": str|None, "synthetic_date": bool, "title": str|None, "text": str}
Usage: python scripts/clean_corpora.py [--raw-dir data/raw] [--out-dir data/clean]
"""
import argparse
import json
import random
import re
from pathlib import Path
MONTHS = ("January February March April May June July August September "
"October November December").split()
MONTH_RE = "|".join(MONTHS)
# ---------------------------------------------------------------- helpers
def gutenberg_body(text: str) -> str:
"""Slice out the Gutenberg header/footer."""
start = re.search(r"\*\*\* ?START OF (THIS|THE) PROJECT GUTENBERG.*?\*\*\*", text)
end = re.search(r"\*\*\* ?END OF (THIS|THE) PROJECT GUTENBERG", text)
s = start.end() if start else 0
e = end.start() if end else len(text)
return text[s:e]
def unwrap(paragraph: str) -> str:
"""Join hard-wrapped lines into a single flowing paragraph."""
return re.sub(r"\s*\n\s*", " ", paragraph).strip()
def paragraphs(text: str) -> list[str]:
return [p for p in re.split(r"\n\s*\n", text) if p.strip()]
def clean_ws(s: str) -> str:
s = re.sub(r"[ \t]{2,}", " ", s)
return s.strip()
def entry_ok(text: str, min_chars: int = 120) -> bool:
"""Reject fragments too short to carry voice."""
return len(text) >= min_chars
# ---------------------------------------------------------------- Pepys
def clean_pepys(raw_dir: Path) -> list[dict]:
text = (raw_dir / "pepys_complete.txt").read_text(encoding="utf-8", errors="replace")
text = gutenberg_body(text).replace("\r\n", "\n")
# Strip indented footnote blocks (Wheatley's notes: lines indented >= 4 spaces).
lines = [ln for ln in text.split("\n") if not re.match(r"^\s{4,}\S", ln)]
text = "\n".join(lines)
# Strip any leftover inline editorial brackets that survived (conservative).
text = re.sub(r"\[[^\[\]]{0,400}?\]", "", text, flags=re.S)
# Drop "ETEXT EDITOR'S BOOKMARKS" blocks if present.
text = re.sub(r"ETEXT EDITOR'S BOOKMARKS.*?(?=\n[A-Z]+ \d{4}|\Z)", "", text, flags=re.S)
month_hdr = re.compile(rf"^({MONTH_RE.upper()})\s+(\d{{4}})(?:-(\d{{2,4}}))?\s*$")
day_start = re.compile(r"^(\d{1,2})(st|nd|rd|th)[.\s]")
entries, cur_month, cur_year, cur = [], None, None, None
for para in paragraphs(text):
first = para.lstrip().split("\n", 1)[0]
mh = month_hdr.match(first.strip())
if mh:
cur_month = mh.group(1).capitalize()
# "1659-1660" style: the latter year is the modern reckoning.
cur_year = mh.group(3) or mh.group(2)
if len(cur_year) == 2:
cur_year = mh.group(2)[:2] + cur_year
continue
ds = day_start.match(para.lstrip())
if ds and cur_month:
if cur:
entries.append(cur)
day = f"{ds.group(1)}{ds.group(2)}"
body = day_start.sub("", para.lstrip(), count=1)
cur = {
"persona": "pepys",
"date": f"{day} {cur_month} {cur_year}",
"synthetic_date": False,
"title": None,
"text": unwrap(body),
}
elif cur:
cur["text"] += "\n\n" + unwrap(para)
if cur:
entries.append(cur)
out = []
for e in entries:
e["text"] = clean_ws(e["text"])
if entry_ok(e["text"]):
out.append(e)
return out
# ---------------------------------------------------------------- Van Gogh
def clean_vangogh(raw_dir: Path) -> list[dict]:
text = (raw_dir / "vangogh_letters.txt").read_text(encoding="utf-8", errors="replace")
text = gutenberg_body(text).replace("\r\n", "\n")
# Drop Ludovici's introductory essay: keep from first letters section onward.
m = re.search(r"^LETTERS TO HIS BROTHER\s*$", text, flags=re.M)
if m:
text = text[m.end():]
# Remove section headings and illustrations.
text = re.sub(r"^LETTERS TO E\. BERNARD\s*$", "", text, flags=re.M)
text = re.sub(r"\[Illustration[^\]]*\]", "", text)
# Letters are separated by asterisk rules.
letters = re.split(r"\n\s*\*(?:\s+\*)+\s*\n", text)
entries = []
for letter in letters:
letter = letter.strip()
if not letter:
continue
# Strip salutation and closing flourishes for diary-ification, but keep body.
letter = re.sub(r"^(MY )?DEAR [A-Z]+,?\s*\n", "", letter)
body = "\n\n".join(unwrap(p) for p in paragraphs(letter))
body = clean_ws(body)
if entry_ok(body, min_chars=200):
entries.append({
"persona": "vangogh",
"date": None, # synthetic dates assigned below
"synthetic_date": True,
"title": None,
"text": body,
})
# The Ludovici letters run roughly 1881 (The Hague) -> 1890 (Auvers), in order.
# Assign evenly spaced synthetic dates across that span so the ghost keeps
# Vincent's chronology of mood (early earnestness -> Arles fever -> late calm).
rng = random.Random(1890)
n = len(entries)
for i, e in enumerate(entries):
year = 1881 + round(i * 9 / max(n - 1, 1))
month = MONTHS[rng.randrange(12)]
day = rng.randrange(1, 29)
e["date"] = f"{day} {month} {year}"
return entries
# ---------------------------------------------------------------- Mansfield
OCR_DIGIT = str.maketrans({"I": "1", "i": "1", "l": "1", "O": "0", "o": "0",
"Q": "9", "q": "9", "p": "9", "g": "9", "S": "5"})
def _ocr_year(line: str):
"""Detect OCR-mangled standalone year headers like 'IQI4' or 'igio' -> 1914/1910."""
s = line.strip()
if not (3 <= len(s) <= 6):
return None
t = re.sub(r"\s", "", s).translate(OCR_DIGIT)
if re.fullmatch(r"19[0-2][0-9]", t):
return int(t)
return None
def clean_mansfield(raw_dir: Path) -> list[dict]:
text = (raw_dir / "mansfield_journal_1927.txt").read_text(encoding="utf-8", errors="replace")
# Trim front matter: start at the first year header (1904) region.
lines = text.split("\n")
body_lines, year, started = [], None, False
page_hdr = re.compile(r"^\s*(Journal\s*(of)?\s*$|Katherine\s*Man.*$|Journal\s+\S{1,8}\s*$)")
for ln in lines:
y = _ocr_year(ln)
if y:
started = True
body_lines.append(f"\n@@YEAR {y}@@\n")
continue
if not started:
continue
if page_hdr.match(ln) and len(ln.strip()) < 30:
continue
body_lines.append(ln)
text = "\n".join(body_lines)
# Remove Murry's editorial bracket blocks (may span lines).
text = re.sub(r"\[[^\[\]]{0,2000}?\]", "", text, flags=re.S)
# Strip OCR page-number artifact lines like " - 102 = " or "102".
text = re.sub(r"^\s*[-=~*•']*\s*\d{1,3}\s*[-=~*•']*\s*$", "", text, flags=re.M)
# OCR de-hyphenation across former line breaks: "to- day" -> "to-day" style joins.
text = re.sub(r"(\w)-\s+(?=\w)", r"\1", text)
# Collapse runs of spaces (OCR double-spacing).
text = re.sub(r"[ \t]{2,}", " ", text)
date_start = re.compile(rf"^({MONTH_RE})\s*(\d{{1,2}})?\s*\.")
entries, cur, cur_year = [], None, 1904
fragment_buf = []
rng = random.Random(1923)
def flush_fragments():
nonlocal fragment_buf
buf, chunk = [], []
size = 0
for p in fragment_buf:
chunk.append(p)
size += len(p)
if size > 700:
buf.append("\n\n".join(chunk)); chunk, size = [], 0
if chunk:
buf.append("\n\n".join(chunk))
for b in buf:
if entry_ok(b, 250):
entries.append({
"persona": "mansfield",
"date": f"{rng.randrange(1,29)} {MONTHS[rng.randrange(12)]} {cur_year}",
"synthetic_date": True, "title": None, "text": clean_ws(b),
})
fragment_buf = []
for para in paragraphs(text):
ym = re.match(r"@@YEAR (\d{4})@@", para.strip())
if ym:
if cur: entries.append(cur); cur = None
flush_fragments()
cur_year = int(ym.group(1))
continue
p = unwrap(para)
dm = date_start.match(p)
if dm:
if cur: entries.append(cur)
flush_fragments()
month, day = dm.group(1), dm.group(2)
date = f"{day} {month} {cur_year}" if day else f"{month} {cur_year}"
cur = {"persona": "mansfield", "date": date, "synthetic_date": False,
"title": None, "text": clean_ws(p[dm.end():])}
elif cur and len(cur["text"]) < 1200:
cur["text"] += "\n\n" + clean_ws(p)
else:
if cur: entries.append(cur); cur = None
fragment_buf.append(clean_ws(p))
if cur: entries.append(cur)
flush_fragments()
return [e for e in entries if entry_ok(e["text"])]
# ---------------------------------------------------------------- MacLane
def clean_maclane(raw_dir: Path) -> list[dict]:
entries = []
# --- The Story of Mary MacLane (1902): right-aligned date headers.
text = (raw_dir / "maclane_story.txt").read_text(encoding="utf-8", errors="replace")
text = gutenberg_body(text).replace("\r\n", "\n")
text = re.sub(r"\[(Photograph|Illustration)[^\]]*\]", "", text)
text = text.replace("_", "").replace("=", "")
date_line = re.compile(rf"^\s{{20,}}({MONTH_RE})\s+(\d{{1,2}})(,\s*(\d{{4}}))?\.?\s*$",
flags=re.M)
cur_year = 1901
marks = list(date_line.finditer(text))
for i, m in enumerate(marks):
if m.group(4):
cur_year = int(m.group(4))
date = f"{m.group(2)} {m.group(1)} {cur_year}"
seg = text[m.end(): marks[i + 1].start() if i + 1 < len(marks) else len(text)]
body = "\n\n".join(unwrap(p) for p in paragraphs(seg))
body = clean_ws(body)
if entry_ok(body):
entries.append({"persona": "maclane", "date": date,
"synthetic_date": False, "title": None, "text": body})
# --- I, Mary MacLane (1917): entries headed by right-aligned day-words
# ("To-day", "To-morrow"), preceded by an italic title line.
text = (raw_dir / "maclane_i.txt").read_text(encoding="utf-8", errors="replace")
text = gutenberg_body(text).replace("\r\n", "\n")
text = re.sub(r"\[(Photograph|Illustration)[^\]]*\]", "", text)
head = re.compile(r"^\s{30,}(To-day|To-morrow|[A-Z][a-z]+day)\s*$", flags=re.M)
title_re = re.compile(r"_([^_\n]{3,80})_\s*$")
marks = list(head.finditer(text))
rng = random.Random(1917)
for i, m in enumerate(marks):
# look back a few lines for the italic title
back = text[max(0, m.start() - 300): m.start()]
tmatch = title_re.search(back.strip())
title = tmatch.group(1).strip() if tmatch else None
seg = text[m.end(): marks[i + 1].start() if i + 1 < len(marks) else len(text)]
seg = seg.replace("_", "")
# cut the next entry's title off the tail
seg = re.sub(r"\n[^\n]{3,80}\n\s*$", "\n", seg)
body = "\n\n".join(unwrap(p) for p in paragraphs(seg))
body = clean_ws(body)
if entry_ok(body):
# The book was written Jan-Mar 1917 in Butte; scatter synthetic dates there.
date = f"{rng.randrange(1,29)} {MONTHS[rng.randrange(0,3)]} 1917"
entries.append({"persona": "maclane", "date": date,
"synthetic_date": True, "title": title, "text": body})
return entries
# ---------------------------------------------------------------- main
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--raw-dir", default="data/raw", type=Path)
ap.add_argument("--out-dir", default="data/clean", type=Path)
args = ap.parse_args()
args.out_dir.mkdir(parents=True, exist_ok=True)
cleaners = {"pepys": clean_pepys, "vangogh": clean_vangogh,
"mansfield": clean_mansfield, "maclane": clean_maclane}
stats = {}
for name, fn in cleaners.items():
entries = fn(args.raw_dir)
out = args.out_dir / f"{name}.jsonl"
with out.open("w") as f:
for e in entries:
f.write(json.dumps(e, ensure_ascii=False) + "\n")
chars = sum(len(e["text"]) for e in entries)
stats[name] = (len(entries), chars)
print(f"[clean] {name:10s} entries={len(entries):5d} chars={chars:9,d} "
f"avg={chars // max(len(entries),1):5d}")
total = sum(c for _, c in stats.values())
print(f"[clean] TOTAL chars={total:,} (~{total // 4:,} tokens)")
if __name__ == "__main__":
main()
|