The largest book in the corpus is a collection of printed web pages, and
every page carries what the browser put at its foot:
http://www.u.arizona.edu/~chalmers/biblio/1.html (54 of 100) [4/7/2002 1:48:08 PM]
That was a reference with a URL and nothing else, one per page, and it is why
that book wanted a model call on all 1,114 of its chunks. An address followed
by which-page-of-how-many and a print time is furniture, not a citation.
`Spade 1980, 12.` and `Trentman 1970; Spade 1980.` are short cites: they point
at a full entry elsewhere in the same bibliography the way `ibid.` points at
the note above. The author and year they do give are kept and the title is not
asked after, because no reading of that line will produce one.
Put a real model behind the entries this could not read and looked at what it
gave back. Almost none of it was knowledge — it was punctuation already on the
page that the parser was walking past:
Goodwin, F.K. and Jamison, K.R. (1990). *Manicdepressive illness* Oxford
University Press, New York.
The title is in asterisks. It was being read as the journal, because the rule
said italics after an author-date opening are the journal — true of `Peirce
(1878). Plain title. *Popular Science Monthly*`, and false here. What decides
it is where the italics sit: at the head of the tail they are the work, after
the title they are what it appeared in.
Five more the model was covering for, all of them on the page:
* `de la Torre, Carlos` opens on a particle and is a name.
* `Tinbergen, n. 1953` has an initial the scan read in lower case.
* `Theories of Nationalism: A Critical Introduction, Basingstoke: Palgrave`
has two colons and the imprint is the last one, not the first.
* `36Armaud Puig i Tarrech` is a note number run into the name.
* `S. Marc Cohen is Professor of Philosophy at...` opens exactly as a
Chicago note does and is a contributor's biography, not a reference.
A title may quote something — `What's the Meaning of "This"?` — so only an
unbalanced quotation mark means the cut landed inside one. A question mark
ends a sentence, which is how that title was being lost. A trailing backslash
is a line break conversion left behind.
`Altmann, M. 1951, A Study of Behaviour. Sociometry 14, 351-354.` names an
author, a year, a journal and a page range, and has no bracketed year, no
`p.`, no `ed.` and no `Press,` in it — so the test chunkmd uses to spot a page
of notes says it cites nothing, and it was being dropped before it was read.
citefind adds its own test alongside, for the same reason it added its own
marker: chunkmd is deciding whether a chunk is a page of notes and can afford
to miss one entry in thirty, and this is deciding whether a reference exists.
Coverage of the model's citations rises to 84% on bibliographies.
The verse pattern now covers the forms a real page of them uses: a chapter and
verse with no book named, a note number glued into the middle of the run, a
chapter range, and a semicolon inside a verse list.
Found by putting a real model at the end of the chain: a page of notes from a
theological book composed a prompt asking who wrote Matthew 4:17, and the
model rightly answered nothing for all twenty-three entries.
Those are scripture references, which is most of what the notes of such a book
are, and they are now not references at all — no record, and no question put
to anybody about them. A journal and its volume look similar on the page and
are still read as what they are.
Also: the stop that belonged to a marker no longer opens the entry after it.
Tried against three of the largest books in the corpus — 2,700 chunks, 44,000
citations the model had already extracted — and two things showed up that the
smaller sample had not.
A chunk that begins part-way through an entry leaves a closing quotation mark
with no opening one, and the title read off it starts mid-word and runs to the
end of the line. A title carrying a stray quote, an escape, or an editor's
credit is now no title, and `unsettled` says so.
The optional fields are written as null rather than left out. A key that comes
and goes is a key every consumer has to guard, and the documented Python use
raised KeyError on the first reference without a page.
Measured against the citations an LLM had already extracted from 40 books of
the same corpus — 19,340 of them over 3,424 chunks — and improved where the
comparison showed a real gap rather than a difference of task.
On pages of notes, where most citations live: year agreement with the model
44% -> 56%, title 18% -> 23%, author 16% -> 24%.
What changed:
- A note listing several works, semicolon-separated, becomes one record per
work. The note number is kept on all of them.
- A full stop is not a separator. It parts the author from the title in every
bibliography entry there is.
- A page of body text with notes at its foot no longer reads the body as the
first note.
- A run carrying a page reference or a semicolon is not reported as a title.
- Six sentences with no imprint, no page and no identifier is a remark about
the argument, not a reference.
- A translator standing where the journal goes is a credit, not a container,
and `(London, 1973)` is a place, not a publisher.
The model is not ground truth and the comparison is not a score: on several
sampled disagreements it named the journal where the article was wanted, or
the collected works instead of the essay.
Cuts a reference list into entries — at the note numbers where there are any,
at the layout where there are not — and reads each as far as its punctuation
allows. Fields that would not parse are named in `unsettled` rather than
guessed at, and `--prompt` asks a model about exactly those and no others.
Over 202 real reference sections from the corpus: 8,566 references, 95% with
a title, 82% with a year, 46% settling every field with no model involved.
A title is refused where the entry was never divided, so no record reports
its own text back as the work it cites.
`ibid.` is flagged and left alone — it points at a work, and the entry does
not say which. A rule of dashes is filled in, because the style defines it as
the author above.
chunkmd gains `note_starts` and `cites_something`: recognising a page of
notes and splitting one are the same knowledge, and two copies of that
pattern would drift.
The hashed bag of words asks only whether two stretches repeat the same
words, and cannot tell that ships and vessels are one subject. An embedding
can. segment_with takes one vector per window and falls back to the words
when it is given none, or the wrong number of them — a caller that embedded
some other recording's windows is asking about that recording.
windows_of gives the text to embed, in the order the vectors come back in,
so nobody has to reimplement the windowing pass to use this.
Bound in all three languages, and on the command line as --windows and
--embeddings, which pipe through emb. The C ABI version goes to 2:
catena_segment_cues gained an argument.
funkpower