Sperberg-McQueen, C. M., Claus Huitfeldt and Yves Marcoux. “Transcriptional implicature.” Presented at Balisage: The Markup Conference 2026, Washington, DC, August 3 - 7, 2026. In Proceedings of Balisage: The Markup Conference 2026. Balisage Series on Markup Technologies, vol. 31 (2026). https://doi.org/10.4242/BalisageVol31.Huitfeldt01.
Balisage: The Markup Conference 2026 August 3 - 7, 2026
Balisage Paper: Transcriptional implicature
C. M. Sperberg-McQueen
Founder and principal
Black Mesa Technologies LLC
C. M. Sperberg-McQueen† 2024 was
the founder and principal of Black Mesa Technologies, a
consultancy specializing in helping memory institutions improve
the long term preservation of and access to the information for
which they are responsible.
He served as editor in chief of the TEI Guidelines from 1988 to 2000, and also
served as co-editor of the World Wide Web Consortium’s XML 1.0 and XML Schema
1.1 specifications.
Claus Huitfeldt works at the Department of Philosophy of the University of
Bergen, Norway. He was founding Director (1990-2000) of the Wittgenstein
Archives at the University of Bergen, for which he developed the text encoding
system MECS as well as the editorial methods for the publication of
Wittgenstein’s Nachlass - The Bergen Electronic Edition (Oxford University
Press, 2000).
Yves Marcoux
Professeur honoraire
École de bibliothéconomie et des sciences de l’information, Université de
Montréal
Yves Marcoux has been a faculty member at EBSI, Université de Montréal, from
1991 to 2018, then adjunct professor and lecturer until 2025. He was mainly
involved in teaching, research, standardization, and international cooperation
activities in the field of document informatics. He has been engaging in
freelance consulting activities since 2020. Prior to his appointment at EBSI,
Dr. Marcoux worked for 10 years in systems maintenance and development, in
Canada, the U.S., and Europe. He obtained his Ph.D. in theoretical computer
science from Université de Montréal in 1991. His main research interests are
intertextual semantics, the design of communication, markup languages, and
digital humanities.
The concept of transcriptional implicature is proposed as a
way of accounting for common practices in transcription while also
making sense of variations in practice. A formal model implicit in
default transcriptional implicature can be elaborated in different
ways to reflect a number of commonalities as well as variations in
transcription practice which have hitherto eluded precise
description in formal terms. We regard the formalization of
transcription as an essential step towards formalizing the
semantics of colloquial XML vocabularies like TEI.
The publication of this paper is meant as a tribute to the late C. Michael
Sperberg-McQueen by the coauthors. In their mind, the paper represents the best
possible graceful conclusion they managed to produce of the work on
transcription they carried out with Michael over a span of nearly two decades (a
large portion of which the third author was absent from the team).
While we by no means consider ourselves in a position to do
justice to the numerous enticing ideas and practical developments
contributed by Michael on the topic, we have felt it important to
at least try our best to make known some of the insights, hopes,
and the enthusiasm he had regarding the possibilities of
formalization for furthering the understanding of theoretical and
practical aspects of transcription.
Whatever good ideas may be found in this account stem in one
way or another from Michael. Only we, however, must bear the
responsibility for the many shortcomings that may remain. Their
number would have been much smaller had we been able, as we
repeatedly found ourselves wishing, to rely on the wit and
benevolence of our friend to keep us from going astray.
Our first formal publication on this topic came out in 2008 [Huitfeldt / Sperberg-McQueen 2008], followed by a Balisage paper in 2010 [Huitfeldt / Marcoux / Sperberg-McQueen 2010]. The notion of
transcriptional implicature is partly based on (and what we saw as
a low-hanging fruit from) ideas from these earlier publications. What is presented
here was more or less finished some time around 2015, and last touched in 2018.[1] However, it was put on the back burner and has never been formally
published until now.[2]
At the time, we regarded transcriptional implicature as only one of the parts of a
more ambitious and comprehensive formal account, or a theory of
the logic, of transcription. Regrettably, we did not manage to
accomplish this quite demanding task before Michael’s untimely death.
We would like to thank participants of a number of conferences
through the years, such as DH2009, Balisage2010, and DH2014, for
their interest, questions, and suggestions.
In particular, we want to thank Michael’s wife, Marian Sperberg-McQueen, for
encouraging the posthumous publication of this paper.
— Claus Huitfeldt & Yves Marcoux
Introduction
The notion of transcriptional implicature
There is hardly any universal transcription practice: for
every generalization we find exceptions. Is everything in the exemplar[3] transcribed? Not when deletions and irrelevant
material are excluded. Does everything in the transcript reproduce
some word or character in the exemplar? Not when line breaks are
marked explicitly with vertical bars, or notes are added.[4] Many scholarly editions account for variations like
these in an explicit statement of transcription practice. Such
statements typically describe deviations from
the usual practice, but rarely the ways in which it
exemplifies usual practice. Common practice
may sometimes be felt to be so obvious that it needs no mention or
explanation.
By transcriptional implicature for a
given community, we mean informally the things, suggested
or entailed by the rules of transcription, that members of that
community may find it unnecessary to mention explicitly.
Different communities of transcription practice have
different sets of tacit assumptions and thus different rules of
transcriptional implicature. Is there a common core of
transcriptional practice shared by all communities? There may be;
it is an empirical question. A definitive answer would require
more detailed studies of a wider variety of communities of
practice than we can currently manage. Our hypothesis, however, is
that there is such a common core, at least in the sense that the
transcriptional implicature of any community of practice can be
described with reference to some common set of rules.
If (as we conjecture) there is such a thing as the transcriptional implicature of
a given community, then the transcriptional practice of any project of that
community could be described by listing the ways in which it
deviates from the transcriptional implicature. Such
deviations can be addition, modification, or withdrawal of rules.
Thus, the rules of the transcriptional implicature are defeasible for particular
projects. They apply, except if explicitly excluded.
If (as we conjecture) the transcriptional implicature of a community can in turn
be described as a set of deviations from some default transcriptional implicature,
then it follows that any project’s transcription practice could be described with
reference to the default transcriptional implicature, by merging the list of ways
in which the community’s implicature deviates from the default implicature with
the list of ways in which the project’s practice deviates from the default
community practice.
The (conjectured) default transcriptional implicature is a set of rules which
apply by default, but which may be overridden in particular cases, analogous to
the rules of conversational implicature proposed by H. P. Grice as a way of
explicating the logic of everyday conversation [Grice 1975].[5]
Earlier work
Since most digital transcriptions are represented today using
markup (e.g., TEI), providing a formal model
of transcription might prove useful for work on the semantics of
markup. Indeed, prevalent approaches to markup semantics are based
on formal languages like first-order predicate logic [Sperberg-McQueen / Huitfeldt / Renear 2001], which would a priori blend
well with a formal model of transcription, yielding a framework in
which meaning can be assigned to markup constructs whose
colloquial definitions involve transcription, for example,
p elements defined as
transcription of a text block in a
manuscript.
Some have proposed to explicate the meaning of markup by
specifying, for each construct in a markup vocabulary, a sentence schema in a
natural language, with blanks to be filled in with data from the document [Marcoux 2006, Marcoux / Rizkallah 2007, Marcoux / Rizkallah 2009]; others
make a similar proposal but allow sentence schemata in formal languages like
first-order predicate logic as well [Sperberg-McQueen / Huitfeldt / Renear 2001, Sperberg-McQueen et al. 2002, Simons et al. 2005, Sperberg-McQueen 2005]. This appears
straightforward, although far from trivial, for metadata [Wickett / Renear 2009]
and perhaps even for born-digital texts, but how shall the meaning of a
p element be formalized in a markup language which defines it as
containing a transcription of a text block in a manuscript?
What does it mean for a document to be a transcription of
another document? Laying a formal ground on which such questions can be tackled,
may be seen as contributions to the semantics of markup, as applied in
transcription.
Early work [Huitfeldt / Sperberg-McQueen 2008] has explored the nature
of the similarity between transcripts and their exemplars. If
documents are defined as sequences of characters, then perhaps the
similarity consists simply in the exemplar and the transcript
containing the same sequence of characters? This can be formalized
but proves disappointing, partly because the definition of
documents as sequences of characters omits text structures like
division into paragraphs and partly because the model offers no
way of describing disagreements among transcribers about how to
read the exemplar, or about which character distinctions (e.g.,
i/j, u/v, s/ſ) to retain and which to level. It is also
wrong for the reason that few transcripts have
exactly the same character sequence as their
exemplar.
Later work has extended the model by introducing boolean
(disjunctive and conjunctive) types, which allowed explicit
representation of ambiguity, as well as
readings, which allowed modeling transcriber
agreement and disagreement explicitly [Sperberg-McQueen / Huitfeldt / Marcoux 2009],
and lifting the analyses from characters to higher-level textual
structures [Huitfeldt / Marcoux / Sperberg-McQueen 2010].
Formalization seems interesting to us for two reasons: First,
and most importantly, formalization encourages a level of
explicitness and rigor which may help us better understand what
transcription is, and thereby also inform our discussions and
judgments in concrete cases. Second, now that transcription is
mostly done digitally, a formal model may help us build better
applications and better understand their possibilities and
limitations.
Our goal is simply to describe formally the relationship that
exists between two documents when the one is said to be a
transcription of the other. It is not our goal
to establish normative criteria which may serve to distinguish good
from bad or right from wrong in transcription, nor to construct a
theory of the psychological or physical processes involved in
transcription.
The term transcription may be used
variously to refer to the act or
process of transcribing a document, to the
physical product of that act (that is, another document), or to
the relation between the two documents.[6]
Our formal approach focuses on the
relation that must obtain between two
documents for the one to be a transcription of the other. Where
such a relation obtains, we call one of the documents a
transcript, and the other an
exemplar.
We take as basic facts about this relation that:
any exemplar antedates its
transcript,
the transcript has been made by someone for certain
reasons and with certain
intentions,
it has been made on the basis of
access to the exemplar (or facsimile of
the exemplar), and
the text of the transcript is the same as, or at least
similar in some way to, the text of the exemplar.[7]
Of all those aspects of the relation between exemplar and
transcript, our formalization addresses only the last one, that
is, the similarity relation that is assumed
to obtain between the texts of a transcript and of its
exemplar.
The kinds, levels or degrees of text similarity required for
a document to be counted as a transcription of another obviously
vary from context to context, and it is not hard to find examples
of disagreement about the matter. The formal model must be
flexible enough to allow for such variations.
The basic notions
The main idea in our approach to formalization is that
documents are textual objects[8] and, as such, can be analyzed in terms of a theory of
types and tokens.[9] In turn, the similarity
relationship that must obtain between two documents for one to be
a transcript of the other (we will call it
T-similarity) is defined in terms of
type-token analyses of the documents.
Here are the basic notions on which the current account is based.
document: A document (in particular, an
exemplar or a transcript) is a
physical phenomenon containing or exhibiting marks
(for example, the copy of Moby Dick on the third author’s nightstand, the
Wittgenstein notebook cataloged as MS 108 in the Wren library).
mark: A mark is a
perceptually discernible arrangement of physical reality in a
document; some (and perhaps, but not necessarily, all) marks
are tokens (for example, words written in
pen or pencil in a manuscript, or the printed words, fly specks
and pencil marks of an old book).[10]
token: A token is a mark
determined to instantiate a particular type while
reading the document (for example, a letter, punctuation mark, a word or a
sentence on a piece of paper or parchment). Tokens can be basic (atomic) or
compound. A compound token is one that is read as the composition of smaller
tokens, called its subtokens. A whole document is viewed as a single
highest-level compound token, i.e., one that is not part of
a larger compound token.
type: A
type is a character, letter, word,
sentence, chapter, or any other object instantiated by one or
more tokens in one or more documents (for example, the
character 'A' of the Latin alphabet, or the sentence 'I know
Verona'). Like tokens, types can be basic or compound. A type
entering in the composition of a compound type is said to be a
subtype of the latter.
For example, the written word "cat" might be read as a
compound token instantiating the compound word type 'cat', and
composed of the three basic subtokens "c", "a", and "t",
instantiating respectively the basic letter types 'c', 'a', and
't', which are subtypes of the word type 'cat'.[11]
Documents (including exemplars and transcripts), which are
highest-level compound tokens, each instantiate a
(highest-level) compound type.
When a token instantiates some given type, we also say the token
is assigned the given type or is
of the given type.
Types can also be conjunctive or
disjunctive, corresponding to ambiguous tokens. For
example, in a manuscript where it is not possible to determine whether some
token is of type 'i' or 'j', yet clear that it is not of any other type, the
token can be said to be of the disjunctive type 'i/j'. In some documents, as
for example, so-called ambigrams, certain tokens are
meant to be read in more than one way. Those tokens can then be regarded as
conjunctive types.[12]
type system: A type system is a set of type
repertoires, which in turn are sets of types. For example, a type system may contain:
a repertoire of characters used in 17th-century English
manuscripts,
a repertoire of words used in the English of that period,
a repertoire of sentences formed using those words, and
repertoires of paragraphs, sections, and other textual structures.
All tokens instantiate at least one type in at least one
repertoire. The same token cannot instantiate more than one type in any
given repertoire. Strictly speaking, a token may represent different types
in different repertoires. For example, the token "I" in English may
instantiate both a character, a word, and a sentence, but not more than one
of each.
There are usually constraints, inherent to the type system, on which
types can be instantiated by a compound token, in terms of the subtypes
instantiated by its subtokens. For example, it could be that any compound
token must be assigned a compound type such that the subtokens of the
compound token instantiate the subtypes of the compound type (as was the
case in the 'cat' example above).
In situations involving a single repertoire, we can speak of the
type (singular) of any given token.
Identification of tokens and types
The process of reading a document, in particular identifying the tokens in the
document and the type that each of them instantiates, is outside the scope of our
model. Whenever we refer to a document, we will allow ourselves to refer also to
its tokens and to the type of each of these tokens, without further
explanation.
Like
most formal models, our work assumes that names can be assigned to
individuals in the universe of discourse, in particular tokens and
types. The practical issues involved in reaching agreement on the
identity of individuals and the names to be used for them are
outside the scope of the formalization. When individuals are as
numerous as the tokens in a novel, of course, agreement on names
for them all is likely to present practical difficulties.
The absence from our account of any description of a method
for assigning names to the tokens of a document reflects not the
belief that such an operation is easy, only the fact that it has
no bearing on the logical structure of transcription, or of
transcriptional implicature. Like the process of reading a
document in the first place, the process of identifying the tokens
in a document sufficiently well to enable productive disagreement
about how to read them is of great practical importance but
outside the focus of our work.
T-similarity
The model defines a relation between documents called
T-similarity. For two documents to be T-similar, they must
be, as high-level tokens, either of the same (high-level) type, or of
similar types, where similarity among types is defined in a
ad hoc manner, in the context of some given transcription
project or set of projects.[13]
T-similarity is the relation that captures
transcription. Its formal definition reflects the
rules of transcription.[14]
In the simplest case, T-similarity is based on type equality;
in other words, exemplar and transcript must be of
identical types, not just similar types.
However, the definition can be relaxed, and this is how we can
hope to capture reasonably realistic aspects of the similarity
between exemplar and transcript in actual projects. Allowing for
such loosening of the similarity conditions is an essential aspect
of our model. We will later further relax the similarity
conditions with the introduction of special
tokens.
The intent behind T-similarity (with suitable versions of
similarity among types) is that it corresponds to the kind of
similarity expected to obtain in practice between an exemplar and
a corresponding transcript, as discussed above. We do not expect,
nor claim, that T-similarity can capture all possible views of
exemplar-transcript similarity, but we suggest it represents a
minimal necessary condition. Also, as already
mentioned, similarity of text is but one aspect of the
relationship between exemplar and transcript in real life:
normally a transcript must follow temporally after its exemplar,
be made in the presence of the exemplar, and be the result of an
intentional effort.
As a mathematical relation, T-similarity is reflexive,
transitive, and symmetric. The irreflexive, intransitive, and
asymmetric nature of the factual relation observed between
transcripts and their exemplars appears to us to be a consequence
of contingent facts and not a property of the similarity relation
holding between the documents.
We have informally defined transcriptional implicature for a
given community as the things, suggested or entailed by the
rules of transcription, that members of that community may find
it unnecessary to mention explicitly. From the point of
view of our formal model, which focuses on the relation between
exemplar and transcript, this amounts to the things
about the T-similarity relation that
go without saying.
Document pairs
We propose to characterize T-similarity as a set of formal statements expressed
in first order logic (FOL), with variables ranging over the types
and tokens of the related documents. Things about the T-similarity
relation (including those that purportedly go without saying) can
then be formally expressed as such statements.
Such a representation of T-similarity can be put to work in various ways. For
instance, given a pair of documents, we can seek to formally establish whether
they stand in a T-similarity relationship or not. This will be done for a simple
pair of documents in [Appendix F].
Or, given one known document (say, a transcript) and assuming that it is
T-similar to some other document (an exemplar), we can ask ourselves what can be
inferred about that other document on the basis of the known document. Exploratory
work in this direction have been described in [Huitfeldt / Sperberg-McQueen 2017].
To do those things, we need to be able to represent pairs of
documents in FOL. The goal of this section is to introduce FOL elements –
predicates and constants – that can be used to say things about
pairs of documents, one exemplar and one transcript, in terms of their types and
tokens.
In the first-order formulae below, we use the following notation: The
variables X, Y, Z, V,, etc. range over all individuals in
the universe of discourse (i.e., the types and tokens of the related
documents). Each individual carries one (and only one) of the predicates:
type(X): X is a type.
e_token(X): X is a token in the
exemplar.
t_token(X): X is a token in the
transcript.
We define the following binary relations (predicates):
typeof(X,Y): token X instantiates type
Y.
transcript(X,Y): token Y in the transcript
corresponds to token X in the exemplar.
exemplar(X,Y): token Y in the exemplar
corresponds to token X in the transcript.
As already mentioned, we are silent on how tokens are identified and assigned
types. We are also silent on how the correspondences between exemplar and
transcript tokens represented by the transcript and
exemplar predicates are established.[15]
Finally, we introduce the following two individual constants:
e: the exemplar (as a single, usually compound, token).
t: the transcript (as a single, usually compound, token).
Note that we have:
e_token(e), and
t_token(t).
It should be clear that facts or hypotheses (to be proved or disproved) about
pairs of documents can be formulated using the above apparatus. In particular, an
exemplar-transcript pair can be entirely represented, in terms of its types and
tokens, as a series of statements using the above predicates together with
additional constants standing for the involved types and tokens. An example is
given in [Appendix F].
As previously mentioned, we do not prescribe nor describe any
method for naming the constants corresponding to types and tokens,
we simply take for granted the use of some adequate naming
scheme.
Axiomatization
The above definitions can be expressed as the following set of axioms:
Axioms
transcript domain: For every X and Y, if X is
the transcript of Y, then X is a t_token and Y is an e_token.
no class overlap: Every individual X is either an
e_token or a t_token or a type.
∀(X)( (e_token(X) ⇔ ( ¬t_token(X) ∧
¬type(X))) ∧
(t_token(X) ⇔ ( ¬e_token(X) ∧ ¬type(X)))
∧
(type(X) ⇔ ( ¬e_token(X) ∧
¬t_token(X))))
In most cases, transcript and exemplar will
naturally be inverses of each other. Likewise, transcript and
exemplar will usually each be single-valued. However, there might
in general be cases where they are not. Thus, we do not include those conditions
as axioms. They will, however, turn out to be part of the default transcriptional
implicature, that is, they will be expected to hold unless
explicitly defeated.
Initial formulation of the default transcriptional
implicature
The initial formulation of the conjectured
default transcriptional implicature which we
present here could be described informally as a transcript
contains the same text as its exemplar. Formally, this
means that the exemplar and the transcript are tokens instantiating
the same type: the transcript re-instantiates the type of the
exemplar. Thus, only documents of equal types
are considered T-similar; this corresponds to the simplest case of
T-similarity presented earlier.
T-similarity could then be formalized in a single
rule: ∃(X) (typeof(e,X) ∧ typeof(t,X)).
In interesting cases, however, the type X (of which e and t are
both tokens) will be a compound type consisting of some structure of smaller
compound types (which in turn consist of smaller ones still), instantiated by a
compound token which similarly consists of smaller tokens.
It is thus more adequate to define T-similarity as the conjunction of five
conditions, as follows:
Default transcriptional implicature definition of
T-similarity
Reciprocity:exemplar and transcript are inverses of each other.
For every X and Y, if X is the
exemplar of Y, then Y is the
transcript of X.
This proposition defines formally what it means for
e and t to be T-similar.
With the axioms given earlier and a FOL representation of any pair of documents,
we can verify whether T-similarity, i.e., the conjunction of the five
rules of reciprocity, single-valued mappings, completeness, purity, and type
identity, comes out as a theorem or not. This is described in more detail
below.
Discussion
It may be worth stressing that the rules just given do not constitute a claim that
every transcript is reciprocal, pure, complete, or thoroughly type identical
to its exemplar. They amount to a claim that these things are true
unless otherwise stated or unless the standards of
a given community of practice dictate otherwise. That is, they express
general assumptions about transcripts which are defeasible in particular cases.
Consider different transcripts of the same exemplar. They may
vary for several reasons. As a first application of our model, let
us use it to describe some of those reasons.
Transcripts may disagree about which of the marks in the
exemplar instantiate types and are thus tokens (one transcript may
read a mark as a decorative pen stroke, the other as a letter in a
word). They may agree on the set of tokens found in the exemplar
but disagree on which types they instantiate. Or they may use
different type systems. One, for example, may distinguish the
allographs i/j, u/v, and ſ/s, while the other treats the
allographs as instantiations of the same type (as they are
instances of the same grapheme).
The use of different type systems can lead to the same kinds
of difference between transcripts as different understandings of
the exemplar; failure to understand the nature of the disagreement
(different reading of the exemplar? or different choice of type
system?) can lead to confusion and acrimony. Analysis of
conflicting transcripts through the common lens of a formal model
might reduce such confusion and acrimony.
We observe that a number of common variations in
transcription practice can be classified according to which rule of
the initial formulation of the default transcriptional implicature
they override.
Some transcripts omit deleted material, extraneous material,
illegible material, or material in specific writing systems
(mathematics, Greek, …). In other words, transcripts are not always
entirely complete.
Some transcripts mark lines with bars, sometimes also adding
line numbers. In some cases omissions may be marked by symbols or
standard phrases ("[Illegible]", etc.) In other words, transcripts
are not always entirely pure.
Some transcripts preserve allographic variations, while others level those
distinctions. For example, some transcripts preserve the distinctions between long
and short "s", or vocalic and consonantal "I" and "u", others do not. This might be
seen as a limitation of the rule of thorough type identity. We
believe it is more natural, however, to assume in such cases that the transcripts
employ different type systems, one in which the allographs are considered tokens of
the same type, and another one in which they are not. In other words, different
transcripts do not necessarily employ identical type systems.
Some transcripts silently expand abbreviations, normalize spelling and correct
slips of the pen. At character level, the rules of completeness and purity seem to
be broken. Even so, such transcripts may observe thorough type identity
at higher level tokens, such as words. In other words, transcripts
are not always entirely type identical through and through.
Expansion of abbreviations in brackets or italics can preserve the rules of
default transcriptional implicature on word and higher levels, but introduces
characters in transcripts which lack corresponding characters in the exemplar.
Again, transcripts are not always entirely pure and do not always preserve entirely
thorough type identity.
In cases of doubt, as in the case of parts of manuscripts which are hard to read
because of wear, damage, or difficulties in handwriting, transcribers tend to
interpret words as correctly spelled and sentences as grammatically well-formed, at
least unless there is evidence to the contrary. This is often referred to as the
principle of charity.[17] This tendency may be accounted for in at least two ways: Either we can
regard the token in e as instantiating a disjunctive type and the token in t as
instantiating one of the unobjectionable disjuncts, or else we can regard the
transcriber as choosing to read the token in e as an instantiation of an
unobjectionable type. In the former case, type identity
must be weakened to allow pairs of the form "x/y"
in e and either "x" or "y" in t (so the types are similar but not identical). In
the latter case, type identity is preserved but the reading of e simply ignores
the fact that the passage in e might plausibly be interpreted differently.
Reformulation of the default transcriptional implicature to permit
exceptions
As we have just seen, in almost all actual transcription projects there are at
least a few exceptions to each of the five rules a-e: material in the exemplar not
transcribed (perhaps because irrelevant to the purpose of the transcript), material
in the transcript not present in the original (if only page numbers and footnotes),
and transcription conventions which transcribe selected tokens with tokens of
similar, not identical type.
The examples given in the next two sections ([section “Example 1: The papers of William Penn” and section “Example 2: Melville’s notes in an edition of Shakespeare”]) show a variety of such exceptions. As these examples will
show, concrete transcription practices can be summarized as a set of exceptions to
the rules of reciprocity, single-valued mappings, purity, completeness, and type
similarity. Since the rules, as given in the preceding section, do not countenance
exceptions, it proves necessary, for reasoning about any actual transcript and its
exemplar, to reformulate those rules to capture the qualification that they each
apply except in special cases.
The formalization we now present shows one way in which the defeasibility of the
simple rules just given can be represented in logical formulae.
We extend the document-pair model with the following predicates and relations:
se_token(X): X is a special exemplar
token.
st_token(X): X is a special transcript
token.
typesimilar(X,Y): types X and
Y are considered similar.
In order to account for the inclusion of special exemplar and special transcript
tokens we modify the typeof domain and the no class
overlap axioms as follows:
Note that in the absence of special tokens and of unequal
types that are typesimilar, the reformulation changes nothing to
T-similarity. Thus, the reformulation does not per
se change the default transcriptional
implicature.
However, the reformulation allows a concise description of any transcription
practice in terms of its deviations from the rules of the
default transcriptional implicature. For any given project’s transcription practice,
we can define the extension of the predicates se_token,
st_token, and typesimilar, and combine them with the
rules just given to provide a basis for inference.
The skeptical reader may object to the explanatory value of this approach: In
general, anything can be described as a deviation from any set of rules. More
specifically, the skeptical reader will have observed that taken strictly, this
reformulation amounts to saying that the properties of completeness, purity, and
type-identity will apply in all cases, except when they do not; the formalization
just given has no way to express the expectation that completeness, purity, and
type-identity are the normal, expected, or usual case, and the cases covered by the
predicates se_token, st_token, and
typesimilar (when it does not coincide with type identity) are
special, unusual, and less frequent cases.
A logic designed to formalize defeasible reasoning would
perhaps capture that distinction better; testing the utility of
such formalisms for reasoning about transcription remains a
desideratum for the future. In the meantime, however, we hope that
this formulation in terms of standard first-order logic will
suffice for the purposes of our argument.
Readers may have been struck by some similarity (pun intended) between our
formulation of the default transcriptional implicature and certain
methods or criteria for identifying or measuring document similarity, such as for
example the so-called Levensthein edit distance between strings of characters.[18] The edit distance between two such strings is measured in terms of the
minimal number of deletions, insertions or substitutions of individual characters
in
the one string that is required in order to make it identical to the other. By
analogy, we might think of special character tokens as deletions, special transcript
tokens as insertions, and type similar tokens as substitutions. The difference,
however, is that while Levensthein similarity is well suited for operations on
character strings, it is less well suited for work on documents with a non-trivial
structure.
Can our formulation of the default
transcriptional implicature be put to empirical test? In principle,
yes. At least we can see whether it makes good sense when applied
to the actual transcriptional practices of a wide range of real
projects. In practice, however, we lack the resources required to
establish any really broad empirical basis. In this paper, we have
limited ourselves to looking at two examples of descriptions of
transcription practice. First we discuss an example drawn from a
U.S.-based historical documentary edition (the papers of William
Penn), then the transcription in a literary edition of some
manuscript notes by Hermann Melville.
Example 1: The papers of William Penn
Statement of practice
In the chapter Editorial Method in [Dunn et al. 1981-1986], the editors provide what we regard as a
statement of practice, stating that:[19]
...In this edition, we aim to print a completely faithful
transcript of each original text, including blemishes and errors.
... In general, our editorial interpolations [enclosed in square
brackets] within the text are minimal...
We observe that the first sentence seems at least implicitly to confirm our rules
of completeness and purity: Every token of e is transcribed by exactly one token
of the same type in t, and every token in t is exemplified by exactly one
token of the same type in e. The second sentence introduces an exception:
transcripts also contain additional material in the form of editorial
interpolations (that is, transcripts are not entirely pure). However, such added
material is explicitly marked by square brackets. The fact that the editors find
it worth pointing this out may be taken to suggest that it does not go without
saying in the relevant community that blemishes and errors are retained, or that
additions are always marked explicitly. The editors continue:
Our editorial rules may be summarized as follows:
1. Each document selected for publication in The
Papers of William Penn is printed in full. ...
At first sight, this looks like a straightforward
confirmation of the rule of completeness. So why does it not go
without saying? Perhaps because it is not altogether common
practice of this community to print every document in full. Or
perhaps because the edition is a selection (that is, not complete
in the sense of containing all the papers of William Penn) the
editors found it important to make clear that although their
transcript is not complete with respect to the entire body of
material, it is complete with respect to each individual
document.
2. Each document is numbered, for convenient
cross-reference, and is supplied with a short title.
From this we can infer that transcription breaks the implicit
rule of purity by adding document numbers
and titles, which do not occur in the exemplar. It seems unlikely
that the editors find it necessary to make this explicit statement
on the assumption that otherwise readers would believe that
William Penn himself numbered his letters and provided them with
titles. It is more likely that the community in question does
expect to find some indication of where one document ends and the
next begins as well as some means of referring to individual
documents, but has not agreed on a uniform way of doing so.
3. The format of each document (including the salutation and
complimentary closure in letters) is rendered as in the original
or copy...[20]
It is not entirely clear from this, or from the rest of the declaration of
practice, what the editors mean by the format of a document.
Comparison of a transcript to a facsimile of its exemplar,[21] however, suggests that what they mean is page layout. At least the
transcripts seem to preserve such features as blocks or lines of text flushed to
the right or left or indented, blank lines, and so on. (We also note that,
elsewhere in the edition, poems are printed as lines of verse.)[22]
It is possible that the editors’ practice is based on an explicit or implicit
distinction between phenomena such as paragraphs, salutations, signatures, date
lines, poems, verse lines, etc. In either case, they seem to be referring to what
we would call compound types, and to try to preserve T-similarity between
transcript and exemplar also in this respect.
It is therefore perhaps striking to observe that the
transcripts contain no indication of line or page breaks[23] in the exemplar, and that the editors do not mention
this fact. Should we consider this omission a violation of the
rule of completeness? (That is, are line and page breaks not
considered tokens of some type?) If so, is the reason why the
editors do not mention this omission that it is part of the
transcriptional implicature of the relevant community of practice?
Or do they simply rely on the appearance of the transcript on the
printed page with regular, running lines, etc. to make it too
obvious to deserve mention that line and page breaks of the
exemplar must be different?
The statement numbered 4 in [Dunn et al. 1981-1986, p. 17]
deals with the rendering of datings according to Julian,
Gregorian, and Quaker calendars, a complicated issue the details
of which we do not go into here.
5. The text of each document is rendered as follows:
a. Spelling is retained as written. Misspelled words are not
marked with an editorial [sic]. If the sense of a word is
obscured through misspelling, its meaning is clarified in a
footnote.
b. Capitalization is retained as written. In
seventeenth-century manuscripts, the capitalization of such
letters as "c," "k," "p," "s," and "w" is often a matter of
judgment, and we cannot claim that our readings are definitive.
Whenever it is clear to us that the initial letter in a sentence
has not been capitalized, it is left lower case.
c. Punctuation and paragraphing are retained as written.
When a sentence is not closed with a period, we have inserted an
extra space.
d. Words or phrases inserted into the text are placed
{within braces}.
e. Words or phrases deleted from the text are crossed
through.
f. Slips of the pen are retained as written, and are not
marked by [sic].
g. Contractions, abbreviations, superscript letters, and
ampersands are retained as written. When a contraction is marked
by a tilde, it is expanded.
5.a suggests that this transcript, by retaining original
spelling, does not deviate from general transcriptional
implicature, but also that it is normal practice in the relevant
community to do so, i.e., by silently normalizing spelling or
marking misspelling with [sic].[24] We take 5.f to mean that the editors make a
distinction between slips of the pen (as in 5.a) from
misspellings, but that they treat them the same way.
The insertion of footnotes to clarify the meaning of
misspelled words mentioned in 5.a, however, does represent an
exception to the default rule of purity, as does the expansion of
contractions marked by tilde mentioned in 5.g. Other deviations
from the rule of purity are identified in 5.c for the insertion of
an extra space when a sentence is not closed with a period, and
5.d for the use of braces to surround editorial insertions into
the text.[25]
5.e simply seems to confirm the default rules of
transcriptional implicature in stating that deleted words or
phrases in the exemplar are crossed through in the transcript.
That the editors are explicit about this may suggest that normal
practice in the community is different. (Perhaps inserting special
markers for deleted text (breaking the rule of purity) or leaving
deleted text out (breaking the rule of completeness)).[26]
Statement 5.b seems to be a paradigmatic application of what
we referred to above as the principle of charity. We take it to be
saying that deciding whether a given character in the manuscript
is uppercase or lowercase requires judgment on the part of the
transcriber.[27]
Thus, to a large extent the statements above explicitly
confirm rules which form part of the default transcriptional
implicature; that they are stated explicitly may suggest that in
the relevant community of practice (or among the expected readers
of the edition), it might be common to deviate from the default
rules in these cases.
h. The thorn is rendered as "th,"[28] and superscript contractions attached to the thorn are brought down
to the line and expanded: as "the," "them," or "that." Our justification for
this procedure is that we no longer have a thorn, and modern readers mistake it
for "y." Likewise, since modern readers do not recognize that "u" and "v" were
used interchangeably in the seventeenth century, we have rendered "u" as "v," or
"v" as "u," whenever appropriate.
i. The £ sign in superscript is rendered as "l."
j. The tailed "p" is expanded into "per," "pro," or "pre,"
as indicated by the rest of the word.
k. The long "s" is presented as a short "s." The double "ff"
is presented as a capital "F."
The statement contained in the first clause of the first sentence of 5.h, i.e.,
rendering the letter "Þ" as two characters, "t" and "h", may seem to deviate
from four of the five rules of our default transcriptional implicature: 1) it
breaks the one-to-one correspondence between the tokens of the exemplar and the
transcript (transcript not single-valued), 2)
there are tokens (thorns) in the exemplar which do not appear in the transcript
(incompleteness), 3) there are tokens ("th"-sequences) in the transcript which do
not occur in the exemplar (impurity), and 4) none of the characters in question
are of the same type (no type similarity). We observe that the statement may
suggest that the normal practice of this community would be to represent "thorn"
with tokens of a distinct type, which would not deviate from default
transcriptional practice at all.
However, if we interpret the statement to the effect that the
project employs a type system in which "Þ" and "th" are
tokens of the same type, the practice is entirely in accordance
with the rules of our default transcriptional implicature. On the
one hand, therefore, such an interpretation seems attractive. On
the other hand, it may seem objectionable: In all other contexts
than where they occur in sequence and stand for a thorn, the
tokens "t" and "h" will still be taken as tokens of two different
types.
To this objection we may answer that it is not at all unusual
for the type of tokens, or even the decision about where to draw
boundaries between tokens, to be context dependent.[29] Also, we do have here a principled justification for
the context-dependent treatment of the thorn: A convention already
exists for representing "Þ" as
"th" in English spelling; "Þ" and "th" are
traditionally pronounced identically by modern readers; and, as
the statement itself argues, modern readers tend to mistake the
thorn for a "y". (The last argument seems to appeal to a principle
of charity with the reader.)
If we are not convinced by these arguments and thus reluctant
to accept that "Þ" and "th" may be regarded as tokens of the
same type, one alternative might be to interpret the first clause
of 5.h to the effect that the thorn is regarded as a contraction
which is expanded to "th". In that case, the practice constitutes
a deviation from default rules on a par with other contractions
— see the discussion of tilde in 5.g above. In any case, the
same goes for the statement made in the second clause of the first
sentence concerning expansion of superscript contractions attached
to the thorn.
Now to the second sentence of 5.h, which observes that "u"
and "v" were used interchangeably in the seventeenth century, and
therefore the one is rendered as the other (or vice versa)
"whenever appropriate" — presumably this means according to
the expectations of modern readers (and thus this practice may be
seen as another instance of "charity to the reader"). This may be
a case for assuming that the project employs a type system
according to which "u" and "v" are tokens of the same type, and in
that case no deviation from default transcriptional implicature is
implied.
What may seem awkward about this analysis is that even if "u"
and "v" are in free distribution in the seventeenth century texts
(alternate realizations of the same type, just as allographs are
alternate realizations of the same grapheme), they are certainly
distinct types (and indeed distinct graphemes) in modern English.
An alternative analysis, therefore, may be that the type system of
the exemplar subsumes that of the transcript: every type in the
transcript corresponds to exactly one type in the exemplar, but
more than one type in the transcript ( here, "u" and
"v") may correspond to the same type in the exemplar
(here, the type instantiated by the tokens we read as
"u" and "v" in the exemplar). Either
way, no deviance from the default transcriptional implicature is
implied.
Statements 5.i and 5.k also describe the type system used:
superscript £ is treated as a token of type "l",[30] and long and short "s" are regarded as tokens of the
same type. This will produce less ambiguity in the transcript than
was the case for "Þ" and "th" or for "v" and "u", since long
"s" will not occur at all in the transcript. Regarding "ff" as a
token of capital "F", however, could in principle lead to
ambiguity.
Statement 5.j may seem to break the rule of purity for the expansion of the
abbreviations mentioned. However, if tailed "p"s are regarded as tokens (depending
on context) of the types "per", "pro", or "pre", the rule is preserved.
The statements 5.h-k may or may not be understood as representing deviations from
the rules of default transcriptional implicature, deviations which are either not
generally shared, or dealt with in other ways, by the community of practice. In
either case, giving a formal account of all details involved here will lead to a
certain amount of complication. As a first simplifying approximation, we may
suggest that the statements imply that the project at hand employs a type system
in which the following tokens are regarded as tokens of the same type:
"Þ" and "th"
"Þ" with superscripts and "the", "them", or "that" (depending on
context and/or the nature of the superscript, — the statement is
silent on this point)
"u" and "v"
"U" and "V"
Superscript "£" and "l"
tailed "p" and "per", "pro", or "pre", as indicated by the rest of the
word.
long and short s
"ff" and "F"
Under this assumption, the statements 5.h-5.k do not represent any
deviation from default transcriptional implicature: reciprocity, single-valued
mappings, purity, completeness and thorough type similarity are all preserved.
l. Words underlined in manuscript are
italicized.
m. Blanks in the manuscript, missing words, and illegible
words are rendered as [blank] or [missing word] or [illegible
word] or [illegible deletion]. If a missing word can be supplied,
it is inserted [within square brackets]. If the supplied word is
conjectural, it is followed by a question mark.
Statement 5.l, that words underlined in the exemplar are
rendered in italics in the transcript, may be seen in different
ways within our framework of rules of transcriptional implicature.
(The following remarks are relevant also to statement 5.e above,
where deleted words or phrases in the exemplar are rendered as
crossed through in the transcript.) We might simply assume that
the editors regard underlined and italicized tokens as typesimilar
(like allographs of the same grapheme). If so, however, why do
they preserve the distinction in the transcript? After all, the
distinction between other allographs of the same grapheme, like
long and short s, are not preserved.
It may seem more natural to assume that the editors
understand underlining in the exemplar as signaling a higher-level
feature, let us call it emphasis, which
pertains not to the atomic letter tokens individually, but to the
entire underlined word, and that they have chosen to signal this
feature by other means, i.e., italics, in the transcript. On this
assumption, an underlined word in the transcript is treated as a
compound token of a type different from the same word without
underlining. Also on this account, however, we might regard this
as a way of preserving type similarity, and thus in accordance
with the rules of transcriptional implicature.
Statement 5.m, that information about blanks, missing words, illegible words, and
conjectural readings are represented by words or phrases within square brackets,
signals a break with the rules of both completeness and purity. Again, we may
interpret the editors’ statement as a confirmation that these rules are otherwise
followed. The reason why they see a need to make this explicit may be either that
these rules are normally followed in such cases, or that the phenomena in question
are normally signaled in other ways.
Interpretation in terms of T-similarity
Given the reformulation of the default transcriptional
implicature offered above, we can interpret the statement of
practice in terms of the three predicates
se_token (special exemplar token),
se_token (special transcript token), and
typesimilar (similar types).
As far as we can tell, the only tokens in e that are not
present in t are the tildes marking contraction. However, these
contractions are expanded in t, so one might argue that this is
a case of type similarity rather than special exemplar
tokens.
The discussion above indicates that the following tokens in
t are special transcript tokens, i.e., that they correspond to
nothing in e:
footnotes (and their markers)
extra space at end of sentence lacking final
punctuation
braces marking inserted material
expansions of contractions (a) marked by tilde, (b) using
thorn + superscripts as the,
them, that, (c)
using tailed “p” as per,
pro, pre.
constituent “t” and “h” of “th” = thorn.
constituent “l” and “.” of “l.” = superscript £.
We have concluded that the type system of t may be assumed
to be the same as that of e, but also noted that the transcript
seems to assume several cases of type similarity. Types t1 and
t2 are similar iff:
t1 = t2
or t1 = thorn, t2 = 〈 “t”, “h” 〉
or t1 = superscript £, t2 = “l”
or t1 = “ff”, t2 = “F”
or (disjunctive(t1) ∧ t2 ∈ disjuncts(t1))
We conclude that together, the statement of practice and our
study of the relation between the exemplar and the transcript, may
confirm our hypothesis that the underlying idea of transcription
in this edition is that of T-similarity.
Example 2: Melville’s notes in an edition of Shakespeare
Statement of practice
On pages 955 to 970 of [Hayford / Parker / Tanselle 1988], the editors discuss notes made by
the American author Herman Melville in an edition of Shakespeare. On pages 967-970
they transcribe the notes and provide facsimile images of the pages. (Facsimiles
of Melville’s notes and the editor’s transcript, taken from [Hayford / Parker / Tanselle 1988], can be found in [Appendix B].) On page 967 they provide a guide to
Symbols used, which we quote in full:[31]
[1] [...] revision or insertion
enclosed in square brackets was made later than initial inscription of
leaf
[2] <...> letters or words
enclosed in diamond brackets were canceled by lining out
[3] <...>word letter(s) or word(s)
written over are enclosed in diamond brackets closed up to the following
word or letter that was superimposed
[4] ?word prefixed by a question mark
indicates conjectural reading
[5] xxxx undeciphered letters (number
of x’s approximates numbers of letters involved)
[6] all words in roman are Melville’s
[7] all words in italics outside
brackets are words Melville underlined
[8] all words in italics inside
brackets are editorial
There is no explicit statement, in this short text, that the exemplar has been
transcribed in full or that nothing appears in the transcript that is not
transcribing something in the exemplar. On the other hand, statement [6], that all
words in roman are Melville’s, and statements [4] and [8], which explain how
editorial conjectures and additions are marked, indicate that great care has been
taken to let the reader know which parts of the document are added by the editors.
One way of making sense of this is to infer that for these transcribers
it goes without saying that unless otherwise indicated, the
transcript is complete and pure in our sense.[32]
Statements [1], [2], [3], and [7]
indicate that the transcribers re-instantiate all insertions,
cancellations, overwritings, and underlined text in the original,
using the notations indicated. We model this by taking insertion,
cancellation, overwriting, and underlining as compound types,
which can like other types be recognized in the exemplar and
re-instantiated in the transcript.[33] These statements have a further implication for the
type system to be used in reading the transcript: the charitable
reader will infer that the various forms of brackets used to
signal insertions, cancellations, and overwriting do not appear in
the exemplar, since if they did, any occurrence of them in the
transcript would become ambiguous.[34]
Statement [5] describes an exception to the usual rule of type-identity between
tokens in e and tokens in t, and also to the default 1:1 mapping between
tokens in the two documents. Undeciphered tokens are transcribed by an
approximate and not an exact number of
x’s, because tokens cannot be counted reliably until they have been identified as
tokens of specific types. (To be a token is to be a token of a specific type.) And
yet, if the occurrences of x in the transcript are tokens,
then they must surely be tokens of some type.
This would call for treating a sequence of n occurrences of
x as a single token of the type undeciphered sequence of letters about n characters
wide. In the exemplar, the undeciphered sequence would
be an atomic (or basic) token, not a compound one; in the
transcript, it would be a compound token composed of an
appropriate number of occurrences of x. The
undeciphered token in the exemplar would then illustrate the
principle that different readers may plausibly read an exemplar in
different ways: a reader who manages to decipher a word will
assign it and its characters to the appropriate types, while a
reader who finds the word illegible will assign it to an undeciphered letter-sequence type of
appropriate length.
(In the formalizations of the exemplar and the transcript below, however, we rely
on a source of information[35] according to which the sequence of letters in question, which is
transcribed as ?almxxxx, has been deciphered as
almanacks, and use the latter in the exemplar. Thus, the exemplar
token almanacks is mapped to the transcript token
almxxxx and the two corresponding types are asserted to be
typesimilar.)
In contrast to the Penn Papers, page and line breaks are
preserved in this edition. We observe that on this point none of
the editions make any explicit statement on their choice. On our
account, this suggests that the transcriptional implicatures of
the two communities of practice are different: In this case it
apparently goes without saying that line breaks are preserved, in
the other it apparently goes without saying that they are not.
It is interesting to observe that the statement of practice in the case of
Melville’s notes is formulated in terms of what the reader will see in t and
what it means, whereas the Penn Paper’s statement of practice starts from
phenomena in e and describes the rules the transcribers have followed in
creating t. The statement of practice for Melville’s notes, that is, specifies
more or less directly what inferences the reader can make about e, given what is
in t, whereas the Penn Papers rules specify more or less directly what the
transcriber must do in t, given what is in e. To use the Penn transcripts to
make inferences about e requires that the rules be applied backwards, so to
speak.
Interpretation in terms of T-similarity
As before, we can interpret the statement of practice in
terms of the three predicates se_token
(special exemplar token), se_token (special
transcript token), and typesimilar (similar
types).
Among tokens present in e and not present in t are marks
(arrows, circles, and carets to mark the desired point of
insertion) which indicate where inserted material belongs. (Note
that these are not mentioned explicitly in the list of symbols,
perhaps because they do not appear as symbols in the transcript;
their existence can be inferred only from the editorial remarks in
the transcript, at line 18 of page 969.) None of these types are
in fact transcribed in t. So, they can all be considered special
exemplar tokens in e, having no corresponding tokens in
t.
Note that the phrase Conversation upon Gabriel, Micheal & Raphel –
gentlemanly &c occurs in line 18 of t, but only below in e. This
is accounted for by the fact that transcript does not necessarily
map consecutive exemplar segments to consecutive transcript segments. It does not
involve special tokens on either side.
The List of Symbols makes clear that italicized words within
angle brackets are editorial, not authorial, i.e., they are special
exemplar tokens.
Similarly, a question mark prefixed to a word whose reading
is uncertain is a special transcript token.
Square brackets, angle brackets, angle brackets with
immediately following text, question marks prefixed to words, and
sequences of the character x are all
identified by the List of Symbols as having special meaning to
which no token in the exemplar directly corresponds, i.e., they are
special transcript tokens.
We may take as a given that in normal cases it will be clear
to a competent reader on inspection whether a token in t is or
is not an instance of one of these types. However, one may also
easily imagine cases in which it is not clear; these are not
different in kind from legibility issues for the reader of t,
both consisting in uncertainty as to the type instantiated by a
token (or uncertainty as to whether a given mark is a token or
not). As already pointed out, the process of reading lies outside
the scope of our work.
There are no clear examples of type similarity between tokens in
e and t. However, see our remarks on a normalized transcript
further below.
Proof strategy
We have now discussed the Melville edition’s statement of practice in terms of
the rules of transcriptional implicature. By doing so we have formulated an
account in more or less concise prose of which parts of e and t should be
assumed special exemplar or transcript tokens, and which types should be assumed
type-similar in order for e and t to be considered T-similar. How can we
establish a formal proof to show that, given these assumptions, t and e are
T-similar?
After all, e and t are concrete, visual objects. They
cannot directly be subjected to the formal procedures required for
such proofs, and definitely not to the digital tools we will use
to test such proofs.
What we can do, however, is to let digital representations
of the documents stand in for e and t (and all the tokens
contained therein), and then perform the formal proof on these
representations.
There are many ways this can be done. The one we have chosen is to create
TEI-XML documents representing t and e. From these XML documents we create one
FOL representation of both documents and of the relations between them. By feeding
this representation alongside our transcriptional implicature axioms to a theorem
prover we check whether T-similarity between the two documents can be proven.
As a sanity test, we also produced HTML visual replicas of e and t from
the XML, which are provided in [Appendix D]. We also produced a HTML
visual replica of another – normalized – transcription (discussed in [section “A normalized transcription”]), which can be seen in [Appendix G].
The theorem prover we have used is Vampire. In
earlier work we have used Alloy and Prolog for
similar tasks. One of the reasons we chose Vampire this time, is
that it operates on fairly standard FOL notation, whereas other
tools use more idiosyncratic notations. Moreover, Vampire is less
apt to crash on the relatively large amounts of data involved in
detailed representations of documents.
Explanation by way of a toy example
An exemplar and a transcript
Let us consider the following artificial example of an
exemplar e:
Figure 1: Exemplar
and the following, equally artificial, transcript of
e, t:
Figure 2: Transcript
In order to justify the plausibility of t, we may assume
that the practice of the relevant community (or of this
particular project) is to include insertions and silently omit
deletions in a different hand (in this case, in red), to
normalize spelling (in this case, of Essexe to
Essex), and to add disambiguating remarks between
square brackets (in this case, [the Earl
of]).
XML representations
Our first step is to create an XML representation of e. We
propose:
<doc>
Elizabeth went <del>with</del> <add>to</add> Essexe
</doc>
Our next step is to create an XML representation of t. We
propose:
<doc>
Elizabeth went to <supplied>the Earl of</supplied> <choice><orig>Essexe</orig><reg>Essex</reg></choice>.
</doc>
We observe that t both adds to and leaves things out of
e, in accordance with the transcribal practice suggested.
<supplied>the Earl
of</supplied>, must be counted as special
transcript tokens. Furthermore, we understand
<choice><orig>Essexe</orig><reg>Essex</reg></choice>
to indicate that the tokens Essexe and
Essex are typesimilar.
When it comes to e,
<del>with</del> must be
considered a special exemplar token in relation to t.
Translation to FOL and proof of T-similarity
Together, the XML representations of e and t, as input
to an appropriately devised XSLT stylesheet (see [Appendix E]), yield the following output:
fof(case_specific_facts, axiom,
typeof(e1, t_Elizabeth) &
typeof(e2, t_went) &
typeof(e3, t_with) & % special
typeof(e4, t_to) &
typeof(e5, t_Essexe) &
typeof(t1, t_Elizabeth) &
typeof(t2, t_went) &
typeof(t3, t_to) &
typeof(t4, t_the) & % special
typeof(t5, t_Earl) & % special
typeof(t6, t_of) & % special
typeof(t7, t_Essex) &
typesimilar(t_Essexe, t_Essex) &
transcript(e1, t1) &
transcript(e2, t2) &
transcript(e4, t3) &
transcript(e5, t7) &
% In our examples, exemplar() is the inverse of transcript(). To avoid defining
% one by one the pairs of exemplar(), we can simply assert reciprocity:
reciprocity &
% Vampire is an open-world-assumption system, so we need to explicitly state that
% there are no other pairs of corresponding tokens. The most efficient way to do
% this in our context is to assert single_valued_mappings:
single_valued_mappings &
% Our examples have one type only per token; let Vampire know about that:
![X,Y]: (typeof(X,Y) =>
~?[Z]:(Z!=Y & typeof(X,Z))) &
$distinct(
e1, e2, e3, e4, e5,
t1, t2, t3, t4, t5, t6, t7,
t_Elizabeth, t_went, t_with, t_to, t_Essexe, t_the, t_Earl, t_of, t_Essex
) &
! [X] : ((e_token(X) <=> (X=e1 | X=e2 | X=e4 | X=e5)) &
(t_token(X) <=> (X=t1 | X=t2 | X=t3 | X=t7)))
).
The Case-specific facts are formulated as an axiom in the
form of one conjunction with many conjuncts, representing the facts of e and
t and the relations between them.
Note that this formalization is based on
word-level tokens and excludes any punctuation and white space. Also, it does
not include any compound token or type, silently assuming that the whole
document is a compound token consisting of the sequence of the basic tokens, and
that its type is the compound type consisting of the sequence of the (basic)
types of the basic tokens. This is also the case of the formalization given for
the Melville example in [Appendix F].
The listing above is in the so-called FOF (First-Order Form)
notation used by Vampire. It is perhaps similar enough to the
standard FOL notation used elsewhere in this paper as it is. For
convenience, however, we include a translation to the usual
notation, and group the various conjuncts into numbered groups
for ease of reference:
In Group 1 the tokens in the exemplar (normal as well as special), named
e1..e5, are associated with their respective types. The names
of the types are simply given as a string consisting of the word string in
question, prefixed with t_. In other words, the token
e1 is of type t_Elisabeth, and so on.
In Group 2 the tokens in the transcript, named t1..t7, are
associated with their respective types, in the same manner.
The statement in Group 3 states that the type named
t_Essexe is similar to the type named
t_Essex.
The statements in Group 4 assigns each of the exemplar tokens in e to their
corresponding transcript tokens in t. It may deserve special attention that
the special exemplar token e3 (of type t_with) and the special
transcript tokens t4, t5, and t6 (of types t_the,
t_Earl, and t_of), though represented as tokens
associated with their respective types, are not mentioned in the Group 4
statements; this is what makes them special. Note that special tokens do not
play any role in T-similarity proofs, because none of the conjuncts comprised in
the T-similarity conjecture involve special tokens.
Group 5 uses one of the conjuncts of the Default transcriptional implicature
as a shortcut to avoid stating the pairs of the exemplar
predicate one by one.
The statements in Groups 6, 7, and 8 are there to deal with the open world
assumption of Vampire, and may be primarily of technical interest.[36]
In Group 6, we state that there are no other pairs of individuals than those
mentioned in Group 4 which stand in a transcript relation to each other, and
that no token-type associations exist except those of Groups 1 and 2.
In Group 7, the $distinct predicate is a special Vampire
predicate that makes sure that each of the constants mentioned refer uniquely,
i.e., e1 names an individual not named by any of the other
constants mentioned, etc.[37]
The statement in Group 8 makes sure that there are no other e_tokens or
t_tokens in the universe of discourse than those listed in the right-hand
clauses of the two biconditionals.
When we add the axioms 1-8 and the definitions a-e of transcriptional
implicature formulated earlier to the axiom of Case-specific
facts above, the theorem prover confirms that e and t are T-similar.[38]
Proof of T-similarity
We have gone through the same steps with the exemplar and transcript [Appendix B] of Melville’s notes as described in the previous section on the
toy example.
We first made XML representations of the exemplar and the
transcript, see [Appendix C]. In both, we used fairly
straightforward TEI encoding.[39]
As a sanity check we made sure we could transform the XML files to HTML which
came as close as we thought feasible to the originals in textual as well as visual
aspects. We present these HTML files in [Appendix D].
The stylesheet briefly mentioned earlier, genFof.xsl, is
reproduced in [Appendix E]. This stylesheet takes the two XML files in
[Appendix C] as input, compares them, and produces the output in
Vampire FOF format, to be found in [Appendix F]. The output of the
stylesheet is the “Case-specific facts axiom”.[40]
In the current form of the stylesheet, special token element names are
hard-coded. One might easily extend the stylesheet so as to give users control
over these, and to distinguish between special exemplar and transcript element
names.
The case-specific facts axiom was then submitted to Vampire, together with a
FOF-representation of the axioms 1-8 and of the definitions a-e, and with
T-similarity as a conjecture.
The output obtained from Vampire (shown at the end of [Appendix F])
confirmed that the conjecture is indeed a theorem, i.e., that the XML
representations of the exemplar and the transcript satisfy the T-similarity
predicate. In other words, that e and t are T-similar.
We believe this to show that our definition of T-similarity,
taking exceptions from default transcriptional implicature into
account, is at least feasible. One such test can of course not be
conclusive. As mentioned earlier, our hypothesis that there is a
common core of assumptions underlying transcription in general is
one that can only be tested empirically. We do not have the
resources necessary for extensive empirical work.
A normalized transcription
However, we have at least made one other, very different,
normalized, transcript of the exemplar [Appendix G].
Unlike the transcript discussed above, this one contains more than one
typesimilar statement: & is declared
type similar to and, &c to etc,
almanacks to almanacs, nonsence to
nonsense, nominee to nomine,
Micheal to Michael, and Raphel to
Raphael.
We put the normalized transcript to the same test. Again, T-similarity as
defined here, with exceptions from default assumptions made explicitly and
formally, shows that also the normalized transcript is T-similar to the exemplar.
See [Appendix G].
This illustrates an important and more general point: Very different
transcripts can be T-similar to one and the same exemplar. (Since T-similarity is
symmetric and transitive, this means that the diplomatic and normalized
transcriptions are also T-similar. We do not prove this.)
Another important point illustrated by this exercise is that the
representation of the exemplar, in the way things are set up here, is not
necessarily (or usually cannot even be) independent of the way the transcript is
represented. For example, a deletion in the exemplar should be represented as a
special transcript token if and only if it is omitted from the transcript.
Discussion
One of the things we think that this project illustrates, is the complexity of
document structures and of the relations between documents. Our formalization has
been on a fairly shallow level. We have succeeded in bringing the axioms of
transcriptional implicature down to roughly a dozen fairly simple FOL statements,
but the formal representation of even the short, less than 300 word Melville
document, requires a conjunct of thousands of statements.
Therefore, we would also like to observe here that Vampire has
impressed us. As mentioned, we have tried to do this work with the
help of other tools in the past, but have had to give up because of
insurmountable difficulties with formalization, user interface, or
speed. This is the first time we have had the occasion to do real
work with formalization and automated theorem proving, even though
still on a fairly modest scale. We realize that some of our work in
Vampire might have been done more efficiently with better knowledge
of its possibilities. Other theorem provers exist, but we are not
presently aware of any that are better suited to the task.
We regard the work presented here as merely a proof of the
concept of formalizing the modeling of documents and document
relations, and of the applicability of the notion of T-similarity
to transcription. Any implementation for real work would
have to use other, more user-friendly and efficient tools, tools
which do not, for example, require thorough knowledge of standard
first order logic.
In [Huitfeldt / Sperberg-McQueen 2008] we presented two models of text:
The Grapheme-sequence Model and the Readings
model. We found the Grapheme-sequence Model
unsatisfactory because of a number limitations:
Texts are not just sequences of graphemes, -- they are usually more complex
graph-like structures with notes, variants, alternate readings, etc.
The analysis of document types must reflect
compositionality and levels of types: there are atomic as well
as boolean and compound types, types at the levels of letters,
words, sentences, paragraphs, etc.
There may be different readings of the same document.
In [Huitfeldt / Marcoux / Sperberg-McQueen 2010] we tried to develop a formal
account without these limitations. However, we never reached a full
working formalization, partly because of the problems with finding
suitable formalization tools mentioned above.
With what is presented here, we are in many ways back with the
Grapheme-sequence Model: We represent documents as
sequences — not even of letters, but of word tokens. (We
could of course have chosen to model on character level in stead.
That would have increased the number of tokens in the model,
without making much interesting difference in principle.) The type
structure is also represented as entirely flat, without levels or
compositionality.
Still, with these limitations, we have a working model which
lends itself to full formalization and formal proof. We can only
hope that someone may some time extend it beyond its current
limitations, e.g., along the lines of [Huitfeldt / Marcoux / Sperberg-McQueen 2010].
Some readers may have been suspicious of our move, in the section on Melville,
from discussing the original exemplar and typescript on paper to discussing
XML-representations of them. It was this move, however, which made us aware of a
(perhaps to others obvious) fact: Our model presupposes that the representation of
the exemplar must reflect what is not recorded in the transcript.[41]
Moreover, we think that this move shows that it might be useful to think in terms
of T-similarity or similar formalizations of similarity between digital documents
in
contexts well beyond transcription and scholarly editing.
Some readers may also find it disturbing that we have treated T-similarity as a
"0 or 1" notion. Surely there must be degrees of T-similarity between 0 and 1? We
agree, and we think there might be various ways of expressing such degrees,
depending e.g., on whether one treats global type identity equal to their number or
to their occurrences, whether one takes only the number or also the distribution of
special tokens into account, and so on. However, in order to introduce such degrees
of similarity one must first decide what counts as degree 1, and we have limited our
task to that.
Originally, this talk had the subtitle a contribution
to markup semantics. We can only hope that it will
be.
Conclusions and Future Work
We have proposed the term transcriptional
implicature to denote the rules of inference which
should govern the interpretation of a transcript in the absence of
explicit statements to the contrary. Informally, the rules of
transcriptional implicature (are intended to) capture those facts
about transcription which are too obvious to need saying. We expect
that the rules of transcriptional implicature vary from community
of practice to community of practice; their accurate identification
will be a matter of the sociology of scholarship, philosophy of
science, and close examination of transcription practice. We
believe that the rules of what we call default
transcriptional implicature provide a plausible example
of what the transcriptional implicature for a given community of
practice might look like, and we have shown how the practice of
some concrete examples can be captured formally in ways which
exhibit clearly both the general rules governing transcription and
the specific practices of the individual project.
We hope that this work can provide a basis for a full model of
the logical structure of transcription, of the formal relation
between exemplar and transcript, and of the ways in which a
transcript can be used to infer facts about its exemplar.
Appendix A. Penn: Facsimile of exemplar and transcript
Facsimiles of the first and last page of a letter from Penn to
Lady Conway from 1675, reproduced on the insides of the front and
back cover of [Dunn et al. 1981-1986]:
Figure 3: Penn, letter to Lady Conway, first page
Figure 4: Penn, letter to Lady Conway, last page
Facsimiles of excerpt from the transcript of the corresponding
pages in [Dunn et al. 1981-1986, pp. 356 and 358]:
Figure 5: Penn, letter to Lady Conway, first page
Figure 6: Penn, letter to Lady Conway, last page
Appendix B. Melville: Facsimile of exemplar and transcript
Although the usual practice in digital transcription is to represent non-ASCII
characters by entities, we include them directly, for ease of reading.
XML representation of the exemplar:
<text>
<body>
<p>
<lb/>A seaman figures in The Canterbury Tales.
<lb/>With <del>a</del>many a tempest had his beard been
<lb/>shook. – <del>S</del><del>Deep g</del> Secret grief is a
<lb/>cannibal of its own heart – <emph>Bacon</emph>.
<lb/>| <emph>Claudia</emph> of the <emph>Appian</emph> family, “I wish
<lb/>| some fight or pestilence would thin out
<lb/>| this crowd.” <emph>Arrogance.</emph>
<lb/><emph>Roast beef in the pulpet.</emph>
</p>
<p>
<lb/>An animal of a man — “do eagles wear
<lb/>spectacles?”— Health. — <emph>Contrast</emph>: an
<lb/>over spiritual man.
<lb/>——
</p>
<p>
<lb/>“Yes, Madam, Cain was a godless froward boy, &
<lb/>Reuben (Gen:49) & Absalom” Many pious men
<lb/>have impious children — (Devil as a Quaker)
<lb/>—————
</p>
<p>
<lb/>A formal compact – Imprimis – First – Second.
<lb/>The aforesaid soul. said soul &c – Duplicates –
<lb/>≡“How was it about the temptation on the
<lb/>hill?” &c – D begs the hero to form
<lb/>one of a <emph>“Society of D’s”</emph> – his name would be weighty
<lb/>&c – Leaves a letter to the D – “My
<lb/><emph>Dear D</emph>”
<floatingText>
<body>
<ab>– ↑ Conversation upon Gabriel, Micheal &
<lb/>Raphel – gentlemanly &c</ab>
</body>
</floatingText>
</p>
<p>
<lb/>“Terra Oblivionis” “Hellites” – At the Astor find him
<lb/>making <sic>almanacks</sic> – going to a ball takes a long
<lb/>time making toilette. – The Doctor’s coach stops
<lb/>the way. – “Do you believe all that stuff?
<lb/>nonsence – the world was never made. – “But Is not
<lb/>this you mentioned <emph>here</emph> – in the scriptures?”
<lb/>Receives visits from the principal d’s – “Gentlemen” &c.
<lb/><emph>Arguments</emph> to persuade – “Would you not rather
<lb/>be below with kings than above with fools?”
</p>
<p>
<lb/>It is better to laugh & not sin than to <del>be</del> weep & be
<lb/>wicked. — Ten loads of coal to burn him. —
<lb/>Brought to the stake — warmed himself by the fire.
</p>
<p>
<lb/>Ego non baptizo te in nominee Patris et
<lb/>Filii et Spiritus Sancti – sed in nomine
<lb/>Diaboli. — Madness is undefinable —
<lb/>It & right reasons extremes of one.
<lb/>–Not the <add place="sup">(black art)</add> Goetic but Theurgic magic —
<lb/>seeks converse with the Intelligence, Power, the
<lb/>Angel.
</p>
</body>
</text>
XML representation of the transcript:
<text>
<front>
<docTitle>
<titlePart>NOTES IN A SHAKESPEARE VOLUME <span>969</span></titlePart>
</docTitle>
</front>
<body>
<fw>[on verso of last leaf of Volume VII, page [524]]</fw>
<p>
<lb n="1"/>A seaman figures in The Canterbury Tales.
<lb n="2"/>With <del>a</del>many a tempest had his beard been
<lb n="3"/>shook. – <del>S</del> <del>Deep g</del> Secret grief is a
<lb n="4"/>cannibal of its own heart – <emph>Bacon</emph>.
<lb n="5"/>| <emph>Claudia</emph> of the <emph>Appian</emph> family, “I wish
<lb n="6"/>| some fight or pestilence would thin out
<lb n="7"/>| this crowd.” <emph>Arrogance.</emph>
<lb n="8"/><emph>Roast beef in the pulpet.</emph>
</p>
<p>
<lb n="9"/>An animal of a man — “do eagles wear
<lb n="10"/>spectacles?”— Health. — <emph>Contrast</emph>: an
<lb n="11"/>over spiritual man.
<lb rend="nonum"/>——
</p>
<p>
<lb n="12"/>“Yes, Madam, Cain was a godless froward boy, &
<lb n="13"/>Reuben (Gen:49) & Absalom” Many pious men
<lb n="14"/>have impious children — (Devil as a Quaker)
<lb rend="nonum"/>—————
</p>
<p>
<lb n="15"/>A formal compact – Imprimis – First – Second.
<lb n="16"/>The aforesaid soul. said soul &c – Duplicates –
<lb n="17"/>≡“How was it about the temptation on the
<lb n="18"/>hill?” &c <supplied rend="[">inserted later below in
lines 21–21b after Dear D” — and circled
<lb rend="nonum"/>with guideline to caret here</supplied>
<floatingText>
<body>
<ab>Conversation upon Gabriel, Micheal &
<lb rend="nonum"/>Raphel – gentlemanly &c</ab>
</body><supplied rend="]"></supplied>
</floatingText> – D begs the hero to form
<lb n="19"/>one of a <emph>“Society of D’s”</emph> – his name would be weighty
<lb n="20"/>&c – Leaves a letter to the D – “My
<lb n="21"/>Dear D” – <supplied rend="[]">later insertion in
lines 21–21b, reported in line 18</supplied>
</p>
<p>
<lb n="22"/>“Terra Oblivionis” “Hellites” – At the Astor find him
<lb n="23"/><unclear>making</unclear>
<choice><orig>almanacks</orig>
<reg><unclear>alm<gap quantity="4" unit="chars"/></unclear></reg></choice>
– going to a ball takes a long
<lb n="24"/>time making toilette. – The Doctor’s coach stops
<lb n="25"/>the way. – “Do you believe all that stuff?
<lb n="26"/>nonsence – the world was never made. –
<supplied rend="[">add </supplied> “But<supplied rend="]" /> Is not
<lb n="27"/>this you mentioned <emph>here</emph> – in the scriptures?”
<lb n="28"/>Receives visits from the principal d’s – “Gentlemen” &c.
<lb n="29"/><emph>Arguments</emph> to persuade – “Would you not rather
<lb n="30"/>be below with kings than above with fools?”
</p>
<p>
<fw>[on recto of last blank leaf of Volume VII, page [523]]</fw>
<lb n="31"/>It is better to laugh & not sin than to
<del>be</del> weep & be
<lb n="32"/>wicked. — Ten loads of coal to burn him. —
<lb n="33"/>Brought to the stake — warmed himself by the fire.</p>
<p>
<lb n="34"/>Ego non baptizo te in nominee Patris et
<lb n="35"/>Filii et Spiritus Sancti – sed in nomine
<lb n="36"/>Diaboli. — Madness is undefinable —
<lb n="37"/>It & right reasons extremes of one.
<lb n="38"/>–Not the
<supplied rend="[">inserted above line with caret below</supplied>
(black art)<supplied rend="]"></supplied> Goetic but Theurgic magic —
<lb n="39"/>seeks converse with the Intelligence, Power, the
<lb n="40"/>Angel.</p>
</body>
</text>
Appendix D. Melville: HTML presentations
HTML presentation of the exemplar:
There are no special tokens in the exemplar, since all exemplar tokens
correspond to some token in the transcript.
The token (“almanacks”) mapped to a transcript token of different – but
typesimilar – type (“almxxxx”) is shown in blue.
Figure 11: HTML presentation of the exemplar
HTML presentation of the transcript:
Special transcript tokens and various marks not in the exemplar are shown in
red.
The token (“almxxxx”) mapped to an exemplar token of different – but
typesimilar – type (“almanacks”) is shown in blue.
Figure 12: HTML presentation of the transcript
Appendix E. genFof stylesheet
<?xml version="1.0" encoding="UTF-8"?>
<xsl:stylesheet version="3.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
xmlns:ym="http://www.marcouxmedias.com"
xmlns:xs="http://www.w3.org/2001/XMLSchema"
xpath-default-namespace="" xml:lang="en-CA"
xml:space="ignore" exclude-result-prefixes="#all">
<xsl:output method="text" indent="no" encoding="UTF-8" />
<!-- For normalized transcription, change to true(): -->
<xsl:param name="normalized" select="false()"/>
<xsl:function name="ym:lastIndexOf" as="xs:integer">
<xsl:param name="str" />
<xsl:param name="car" />
<xsl:sequence select="if (contains($str, $car)) then
max(for $i in (1 to string-length($str)) return
if (substring($str, $i, 1) = $car) then $i else 0)
else 0" />
</xsl:function>
<xsl:function name="ym:escAmp" as="xs:string">
<xsl:param name="str" as="xs:string" />
<xsl:value-of select="replace($str, '&', 'et_')"/>
</xsl:function>
<xsl:variable name="fpNoExt">
<xsl:variable name="temp" select="base-uri(/)" />
<xsl:value-of select="substring($temp, 1, ym:lastIndexOf($temp, '.'))" />
</xsl:variable>
<xsl:variable name="indent" select="' '"/>
<xsl:function name="ym:tokenizePlus" as="item()*">
<xsl:param name="text" as="node()" />
<xsl:param name="withOrig" as="xs:boolean" />
<xsl:choose>
<xsl:when test="not($text/ancestor::orig) or $withOrig">
<!-- The [.] in the following is to get rid of possible empty strings at the
beginning and end of the string: -->
<xsl:variable name="toks" select="tokenize($text,'[^a-zA-Z0-9&]+')[.]"/>
<xsl:variable name="toks" select="
let $maybeGap :=
$text/following-sibling::node()[1][self::gap][@quantity] return (
if ($maybeGap) then (
let $fill := string-join(for $i in (1 to $maybeGap/@quantity) return 'x')
return (remove($toks, count($toks)), $toks[count($toks)] || '_' || $fill)
) else $toks)
"/>
<xsl:choose>
<!-- The values of test specified below are project-specific -->
<xsl:when test="$text/ancestor::front
or $text/ancestor::fw
or $text/ancestor::supplied
or $text/ancestor::floatingText
or $text/ancestor::secl
or ($text/ancestor::del and $normalized)
or ($text/ancestor::sic and $normalized)">
<xsl:sequence select="for $t in $toks return '*' || $t" />
</xsl:when>
<xsl:otherwise>
<xsl:sequence select="$toks" />
</xsl:otherwise>
</xsl:choose>
</xsl:when>
<xsl:otherwise />
</xsl:choose>
</xsl:function>
<xsl:function name="ym:addEffSeqNum" as="item()*">
<xsl:param name="toks" as="item()*" />
<xsl:sequence select="
for $n in 1 to count($toks) return
(if (not(starts-with($toks[$n],'*'))) then
count($toks[position() lt $n and not(starts-with(.,'*'))]) || '*' else '')
|| $toks[$n]
" />
</xsl:function>
<xsl:template match="/">
<xsl:variable name="doc1" select="/*"/>
<xsl:variable name="toks1" select=
"ym:addEffSeqNum($doc1//text()/ym:tokenizePlus(., true()))" />
<xsl:variable name="doc2" select="document($fpNoExt || 'transcript.xml')/*"/>
<xsl:variable name="toks2" select=
"ym:addEffSeqNum($doc2//text()/ym:tokenizePlus(., false()))" />
<xsl:variable name="nNormToks" select="count($toks1[not(starts-with(.,'*'))])" />
<xsl:if test="$nNormToks ne count($toks2[not(starts-with(.,'*'))])">
<xsl:message terminate="no"
select="string-join(($nNormToks, count($toks2[not(starts-with(.,'*'))]), ''), ' ')"
>WARNING: Normal token counts differ between documents.</xsl:message>
</xsl:if>
<xsl:text>fof(case_specific_facts, axiom,
</xsl:text>
<xsl:call-template name="processDoc">
<xsl:with-param name="docPrefix" select="'e'" />
<xsl:with-param name="toks" select="$toks1" />
</xsl:call-template>
<xsl:call-template name="processDoc">
<xsl:with-param name="docPrefix" select="'t'" />
<xsl:with-param name="toks" select="$toks2" />
</xsl:call-template>
<xsl:for-each select="$doc2//choice">
<xsl:variable name="toksOrig" select="ym:tokenizePlus(orig//text(), true())"/>
<xsl:variable name="toksReg" select="ym:tokenizePlus(reg//text(), false())"/>
<xsl:if test="count($toksOrig) ne count($toksReg) or not(count($toksOrig))">
<xsl:message select="string-join((count($toksOrig), count($toksReg), ''), ' ')"
>WARNING: Token count mismatch within a <choice> element.</xsl:message>
</xsl:if>
<xsl:for-each select="1 to count($toksOrig)">
<xsl:variable name="toksOrig2" select="let $t := $toksOrig[current()] return
if (starts-with($t,'*')) then substring($t,2) else $t"/>
<xsl:variable name="toksReg2" select="let $t := $toksReg[current()] return
if (starts-with($t,'*')) then substring($t,2) else $t"/>
<xsl:if test="$toksOrig2 ne $toksReg2">
<xsl:value-of select="$indent" />
<xsl:text>typesimilar(t_</xsl:text>
<xsl:value-of select="ym:escAmp($toksOrig2)" />
<xsl:text>, t_</xsl:text>
<xsl:value-of select="ym:escAmp($toksReg2)" />
<xsl:text>) &
</xsl:text>
</xsl:if>
</xsl:for-each>
</xsl:for-each>
<xsl:for-each select="1 to count($toks1)">
<xsl:if test="not(starts-with($toks1[current()],'*'))">
<xsl:variable name="idNormTok" select="substring-before($toks1[current()],'*')" />
<xsl:value-of select="$indent" />
<xsl:text>transcript(e</xsl:text>
<xsl:value-of select="." />
<xsl:text>, t</xsl:text>
<xsl:value-of select="max(
for $i in 1 to count($toks2) return
if (starts-with($toks2[$i], $idNormTok || '*')) then $i else 0
)" />
<xsl:text>) &
</xsl:text>
</xsl:if>
</xsl:for-each>
<xsl:text><![CDATA[
% In our examples, exemplar() is the inverse of transcript(). To avoid defining
% one by one the pairs of exemplar(), we can simply assert reciprocity:
reciprocity &
% Vampire is an open-world-assumption system, so we need to explicitly state that
% there are no other pairs of corresponding tokens. The most efficient way to do
% this in our context is to assert single_valued_mappings:
single_valued_mappings &
% Our examples have one type only per token; let Vampire know about that:
![X,Y]: (typeof(X,Y) =>
~?[Z]:(Z!=Y & typeof(X,Z))) &
]]>
</xsl:text>
<xsl:value-of select="$indent" />
<xsl:text>$distinct(
</xsl:text><xsl:value-of select="$indent" /><xsl:value-of select="$indent" />
<xsl:value-of select="
string-join(for $i in 1 to count($toks1) return ('e' || $i), ', ')
" />
<xsl:text>,
</xsl:text><xsl:value-of select="$indent" /><xsl:value-of select="$indent" />
<xsl:value-of select="
string-join(for $i in 1 to count($toks2) return ('t' || $i), ', ')
" />
<xsl:text>,
</xsl:text><xsl:value-of select="$indent" /><xsl:value-of select="$indent" />
<xsl:variable name="allTypes" select="
(for $i in 1 to count($toks1) return 't_' || ym:escAmp(substring-after($toks1[$i], '*'))),
(for $i in 1 to count($toks2) return 't_' || ym:escAmp(substring-after($toks2[$i], '*')))
"/>
<xsl:value-of select="string-join(distinct-values($allTypes), ', ')" />
<xsl:text>
</xsl:text>
<xsl:value-of select="$indent" />
<xsl:text>) &
</xsl:text>
<xsl:value-of select="$indent" />
<xsl:text>! [X] : ((e_token(X) <=> (X=e</xsl:text>
<xsl:value-of select="string-join(
(1 to count($toks1))[let $i := . return not(starts-with($toks1[$i],'*'))],
' | X=e')" />
<xsl:text>)) &
</xsl:text>
<xsl:value-of select="$indent" />
<xsl:text> (t_token(X) <=> (X=t</xsl:text>
<xsl:value-of select="string-join(
(1 to count($toks2))[let $i := . return not(starts-with($toks2[$i],'*'))],
' | X=t')" />
<xsl:text>)))
).</xsl:text>
</xsl:template>
<xsl:template name="processDoc">
<xsl:param name="toks" />
<xsl:param name="docPrefix" />
<xsl:for-each select="$toks">
<xsl:value-of select="$indent" />
<xsl:text>typeof(</xsl:text>
<xsl:value-of select="$docPrefix"/>
<xsl:value-of select="position()"/>
<xsl:text>, t_</xsl:text>
<xsl:value-of select="ym:escAmp(substring-after(.,'*'))" />
<xsl:text>) &</xsl:text>
<xsl:value-of select="if (starts-with(.,'*')) then ' % special' else ''"/>
<xsl:text>
</xsl:text>
</xsl:for-each>
</xsl:template>
</xsl:stylesheet>
Appendix F. Melville: FOF representation of the document pair (exemplar and
transcript)
The output from the genFof stylesheet (the section entitled
“Case-specific axiom” below) was retouched by hand to account for the short passage
appearing not at the same place in the exemplar and in the transcript.
The output from Vampire, confirming that the conjecture is indeed a theorem,
is:
SZS status Theorem for Exceptional
------------------------------
Version: Vampire 4.9 (commit )
Termination reason: Refutation
Memory used [KB]: 87569
Time elapsed: 5.557 s
------------------------------
------------------------------
Success in time 21.556 s
On a modern laptop computer, Vampire ran for a little less than 22 seconds
trying various proof strategies, of which the first successful one took a little
less than 6 seconds.
Appendix G. Melville: XML, HTML, and FOF for normalized transcript
XML representation of exemplar for normalized transcript:
<text>
<body>
<p>
<lb/>A seaman figures in The Canterbury Tales.
<lb/>With <del>a</del>many a tempest had his beard been
<lb/>shook. – <del>S</del><del>Deep g</del> Secret grief is a
<lb/>cannibal of its own heart – <emph>Bacon</emph>.
<lb/>| <emph>Claudia</emph> of the <emph>Appian</emph> family, “I wish
<lb/>| some fight or pestilence would thin out
<lb/>| this crowd.” <emph>Arrogance.</emph>
<lb/><emph>Roast beef in the pulpet.</emph>
</p>
<p>
<lb/>An animal of a man — “do eagles wear
<lb/>spectacles?”— Health. — <emph>Contrast</emph>: an
<lb/>over spiritual man.
<lb/>——
</p>
<p>
<lb/>“Yes, Madam, Cain was a godless froward boy, <sic>&</sic>
<lb/>Reuben (Gen:49) <sic>&</sic> Absalom” Many pious men
<lb/>have impious children — (Devil as a Quaker)
<lb/>—————
</p>
<p>
<lb/>A formal compact – Imprimis – First – Second.
<lb/>The aforesaid soul<secl>. said soul</secl> <sic>&c</sic> – Duplicates –
<lb/>≡“How was it about the temptation on the
<lb/>hill?” <sic>&c</sic> – D begs the hero to form
<lb/>one of a <emph>“Society of D’s”</emph> – his name would be weighty
<lb/><sic>&c</sic> – Leaves a letter to the D – “My
<lb/><emph>Dear D</emph>”
<floatingText>
<body>
<ab>– ↑ Conversation upon Gabriel, <sic>Micheal</sic> <sic>&</sic>
<lb/><sic>Raphel</sic> – gentlemanly <sic>&c</sic></ab>
</body>
</floatingText>
</p>
<p>
<lb/>“Terra Oblivionis” “Hellites” – At the Astor find him
<lb/>making <sic>almanacks</sic> – going to a ball takes a long
<lb/>time making toilette. – The Doctor’s coach stops
<lb/>the way. – “Do you believe all that stuff?
<lb/><sic>nonsence</sic> – the world was never made. – “But <sic>Is</sic> not
<lb/>this you mentioned <emph>here</emph> – in the scriptures?”
<lb/>Receives visits from the principal d’s – “Gentlemen” <sic>&c</sic>.
<lb/><emph>Arguments</emph> to persuade – “Would you not rather
<lb/>be below with kings than above with fools?”
</p>
<p>
<lb/>It is better to laugh <sic>&</sic> not sin than to <del>be</del> weep <sic>&</sic> be
<lb/>wicked. — Ten loads of coal to burn him. —
<lb/>Brought to the stake — warmed himself by the fire.
</p>
<p>
<lb/>Ego non baptizo te in <sic>nominee</sic> Patris et
<lb/>Filii et Spiritus Sancti – sed in nomine
<lb/>Diaboli. — Madness is undefinable —
<lb/>It <sic>&</sic> right reasons extremes of one.
<lb/>–Not the <add place="sup">(black art)</add> Goetic but Theurgic magic —
<lb/>seeks converse with the Intelligence, Power, the
<lb/>Angel.
</p>
</body>
</text>
XML representation of normalized transcript:
<text>
<body>
<p>
<lb/>A seaman figures in The Canterbury Tales.
<lb/>With many a tempest had his beard been
<lb/>shook. – Secret grief is a
<lb/>cannibal of its own heart – <emph>Bacon</emph>.
<lb/><emph>Claudia</emph> of the <emph>Appian</emph> family, “I wish
<lb/>some fight or pestilence would thin out
<lb/>this crowd.” <emph>Arrogance.</emph>
<lb/><emph>Roast beef in the pulpet.</emph>
</p>
<p>
<lb/>An animal of a man — “do eagles wear
<lb/>spectacles?”— Health. — <emph>Contrast</emph>: an
<lb/>over spiritual man.
</p>
<p>
<lb/>“Yes, Madam, Cain was a godless froward boy, <choice>
<orig>&</orig>
<reg>and</reg>
</choice>
<lb/>Reuben (Gen:49) <reg>and</reg> Absalom” Many pious men
<lb/>have impious children — (Devil as a Quaker)
<lb/>
</p>
<p>
<lb/>A formal compact – Imprimis – First – Second.
<lb/>The aforesaid soul <choice>
<orig>&c</orig>
<reg>etc.</reg>
</choice> – Duplicates –
<lb/>“How was it about the temptation on the
<lb/>hill?” <reg>etc.</reg> <supplied>Conversation upon Gabriel, <choice>
<orig>Micheal</orig>
<reg>Michael</reg>
</choice> <reg>and</reg>
<lb/><choice>
<orig>Raphel</orig>
<reg>Raphael</reg>
</choice> – gentlemanly <reg>etc.</reg></supplied> – D begs the hero to form
<lb/>one of a <emph> “Society of D’s” </emph> – his name would be weighty
<lb/><reg>etc.</reg> – Leaves a letter to the D – “My
<lb/><emph>Dear D</emph>” –
</p>
<p>
<lb/>“Terra Oblivionis” “Hellites” – At the Astor find him
<lb/>making <choice><orig>almanacks</orig><reg>almanacs</reg>
</choice> – going to a ball takes a long
<lb/>time making toilette. – The Doctor’s coach stops
<lb/>the way. – “Do you believe all that stuff?
<lb/><choice><orig>nonsence</orig><reg>nonsense</reg></choice> –
the world was never made. – “But
<choice><orig>Is</orig><reg>is</reg></choice> not
<lb/>this you mentioned <emph>here</emph> – in the scriptures?”
<lb/>Receives visits from the principal d’s –
“Gentlemen” <reg>etc.</reg>.
<lb/><emph>Arguments</emph> to persuade – “Would you not rather
<lb/>be below with kings than above with fools?”
</p>
<p>
<lb/>It is better to laugh <reg>and</reg> not sin than to weep <reg>and</reg> be
<lb/>wicked. — Ten loads of coal to burn him. —
<lb/>Brought to the stake — warmed himself by the fire.
</p>
<p>
<lb/>Ego non baptizo te in
<choice><orig>nominee</orig><reg>nomine</reg></choice> Patris et
<lb/>Filii et Spiritus Sancti – sed in nomine
<lb/>Diaboli. — Madness is undefinable —
<lb/>It <reg>and</reg> right reasons extremes of one.
<lb/>–Not the (black art) Goetic but Theurgic magic —
<lb/>seeks converse with the Intelligence, Power, the
<lb/>Angel.
</p>
</body>
</text>
HTML presentation of exemplar for normalized transcript:
Special exemplar tokens are shown in red.
Tokens mapped to a transcript token of different – but
typesimilar – type are shown in blue.
Figure 13: HTML presentation of exemplar for normalized transcript
HTML presentation of normalized transcript:
There are no special transcript tokens.
Tokens mapped to an exemplar token of different – but
typesimilar – type are shown in blue.
Figure 14: HTML presentation of normalized transcript
FOF representation of the document pair (exemplar and normalized
transcript):
Only the Case-specific facts axiom section is shown, all
other sections are identical to those found in Appendix F.
The output from the genFof stylesheet was retouched by hand to
account for the short passage appearing not at the same place in the exemplar and
in
the transcript.
Again, the output of Vampire confirms that the conjecture is indeed a
theorem:
SZS status Theorem for Exceptional
------------------------------
Version: Vampire 4.9 (commit )
Termination reason: Refutation
Memory used [KB]: 87587
Time elapsed: 5.393 s
------------------------------
------------------------------
Success in time 21.355 s
References
[Carter 1952] Carter, Clarence E.
Historical editing. Bulletins of the national archives,
Number 7. Washington, DC: National Archives and Records Service, August 1952.
National Archives publication number 53-4.
[Dunn et al. 1981-1986] Dunn, Mary Maples et al.
The Papers of William Penn, 5 vols. Philadelphia: University
of Pennsylvania Press, 1981-1986.
[Goodman 1976] Goodman, Nelson.
Languages of Art. Hackett Publishing, 1976.
[Grice 1975] Grice, H.P.
Logic and Conversation. In Syntax and
Semantics, edited by P. Cole and J. Morgan, vol.3,
Speech Acts. New York: Academic Press, 1975.
Reprinted as chapter 2 of his Studies in the Way of
Words. Cambridge, Mass.: Harvard University Press,
1989, pp. 22–40.
[Hayford / Parker / Tanselle 1988] Hayford,
Harrison, Hershel Parker, and G. Thomas Tanselle, ed. Moby Dick, or, The
Whale. Vol. 7 of The Writings of Herman Melville The
Northwestern–Newberry Edition. Evanston [Ill.]: Northwestern University Press;
Chicago: Newberry Library, 1988, rpt. 1994, 1997.
[Robinson / Solopova 2006] Robinson, Peter, and
Elizabeth Solopova. Guidelines for Transcription of the Manuscripts of The
Wife of Bath’s Prologue. 18 March 2006. On the Web
at zenodo.org, accessed October 2026.
[Simons et al. 2005] Simons, Gary, Scott O.
Farrar, Brian Fitzsimons, William D. Lewis, D. Terence Langendoen, and Hector
Gonzalez. The semantics of markup: Mapping legacy markup schemas to a common
semantics. In Proceedings of the 4th workshop on NLP and XML
(NLPXML-2004): held in cooperation with ACL- 04 Barcelona, Spain, pp.
25-32. doi:https://doi.org/10.3115/1621066.1621070. On the Web at [http://www.aclweb.org/anthology/W/W04/W04-0604.pdf], accessed October 2026.
[Sperberg-McQueen et al. 2002] Sperberg-McQueen,
C. M., David Dubin, Claus Huitfeldt, and Allen Renear. Drawing inferences on
the basis of markup. Paper given at Extreme Markup Languages®,
Montréal, 2002. Proceedings of Extreme Markup Languages®
2002. On the Web at
wwp.northeastern.edu (PDF) and www.semanticscholar.org.
[Sperberg-McQueen / Marcoux / Huitfeldt 2018] Sperberg-McQueen, C. M., Yves Marcoux, and Claus Huitfeldt. Transcriptional
implicature: A contribution to markup semantics.Unpublished, 2018. [XML / HTML], accessed October 2026.
[Stevens / Burg 1997] Stevens, Michael E.
and Steven B. Burg. Editing Historical Documents: A Handbook of
Practice. Walnut Creek, London, New Delhi: Alta Mira Press,
1997.
[Tanselle 1989] Tanselle, G. Thomas.
A Rationale of Textual Criticism. Philadelphia: University of
Pennsylvania Press, 1989. 104 pp.
[Vander Meulen / Tanselle 1999] Vander Meulen,
David, and G. Thomas Tanselle. A system of manuscript transcription.Studies in Bibliography 52 (1999), pp. 201-212.
[2] Except for abstracts and slides from various conferences, cf. below.
[3] We use the English term exemplar to
denote the document from or of which a transcript is made, in
preference to the term original, which
evokes confusing associations when the exemplar is itself a
transcript. In our usage, exemplar thus corresponds to what is
referred to in German as the Vorlage or in
French as the antigraphe.
[5] Our notion of transcriptional implicature is indeed inspired by Grice’s
theory, but we make no claim as to the similarities between our notion and
Grice’s theory of conversational and conventional implicature.
[6] In other contexts, such as linguistics, music, or genetics,
transcription refers to different, though
related phenomena — see [Huitfeldt / Sperberg-McQueen 2008, pp
295-96].
[8] Many documents do of course contain non-textual material.
It may be argued that at least some kinds of non-textual
material lend themselves to a type-token analysis, but we do not
believe that always to be the case. Our account is limited to
the aspects of documents which do lend themselves to such
analysis, and we do not argue that all aspects of documents
do.
[9] The presupposition that exemplar as well as transcript lend
themselves to analysis in terms of the concepts of tokens and
types presented above implies that they must be
notations as Goodman defines that
term.
According to Goodman, the requirements of notational
schemes are:
...[A] character in a notation is an abstraction
class of character-indifference among inscriptions. As a
result, no mark may belong to more than one
character. [Goodman 1976, pp. 132-3]. In
our terminology: No token is an instantiation of more than
one type.
...the characters be finitely differentiated, or
articulate. [Goodman 1976, p. 135]. That
is, for any mark it must be at least theoretically possible
to determine whether or not it belongs to a certain sign
(character). In our terminology: For any token, it must be at
least theoretically possible to determine whether it
instantiates a certain type or not.
These are what Goodman calls syntactic
requirements that notational schemes have
to satisfy. In order for a notational scheme to qualify as a
notational system, it must also satisfy
three semantic requirements: It must be
unambiguous,disjoint and finitely
differentiated.
More precisely:
Every notational system must be unambiguous. In other
words, the extension (compliance-class) of a sign (character
or inscription) must be the same for all occurrences of that
sign – it cannot vary from case to case [Goodman 1976, p. 148].
Every extension (compliance-class) must be distinct
(disjoint) from every other extension. In other words, no
object in a domain can belong to two extensions [Goodman 1976, p. 150].
Every extension (compliance-class) must be finitely
differentiated. In other words, for any object in the domain
it must be at least theoretically possible to determine
whether or not it belongs to a certain extension [Goodman 1976, p. 152].
It would be interesting to investigate to what extent our
account of transcription implies that transcription is
notational not only in terms of the syntactic, but also in terms
of the semantic requirements. We think this may be the case, but
do not further investigate the issue as it is not of direct
relevance to the tasks we have set ourselves here.
[10] The reader will note that the notion of mark is very vague;
in view of the many different ways in which written messages can be
constructed or conveyed, we believe this to be unavoidable. The notion
is not defined formally in the model.
[11] When the distinction is not clear from context, we use
single quotes for types and double quotes for tokens.
[13] T-similarity, however, involves more than just type similarity. It also
depends on how the documents are broken down into tokens, and on how the
tokens of one document are mapped to those of the other document (mappings
called transcript and exemplar below).
The reader may feel free to read the T in
T-similarity to stand for text or
transcriptional.
[14] The reader may wonder how rules can be represented by a mathematical relation.
The name rules suggests imperative statements such as
when a sentence is not closed with a period, insert an extra
space or whenever appropriate, render "u" as "v", or "v" as
"u", more than it suggests a relationship. The answer is that
T-similarity represents transcription rules
in a descriptive manner. It specifies how
exemplar and transcript are to stand relative to each other once the
transcription is completed, the implicit rules being: do whatever it
takes for that relation to hold when you are done.
Interestingly, in the statement of
practice accompanying some transcripts, the rules of
transcription are also formulated in descriptive
rather than imperative style: e.g., spelling is retained
as written or the long "s" is presented as a
short "s", rather than retain spelling as
written or convert any long "s" to a short
"s".
[16] The left-hand side of the ⇒ needs to include both
transcript and exemplar to cover cases that
do not satisfy reciprocity.
[17] One might regard this as a principle of giving the author the benefit of the
doubt. Or perhaps one might just as well regard it as an act of charity to the
reader: Unless there is clear evidence of error, there is no reason to bother
the reader with the mere possibility of error.
[19] Here and in the rest of this section, indented quotations
are quotations from pp. 15-18 in [Dunn et al. 1981-1986].
[20] The quotation continues: , with the following two
exceptions. Endorsements are treated as dockets, and entered
into the provenance note (see below). If a document is
undated, an initial date line is supplied [within square
brackets]. If a document is dated at the close but not at the
opening, an initial date line is supplied [within square
brackets], and the closing date line is
retained.
[21] Unfortunately we have only had access in facsimile to the first and the
last page of one letter, which are reproduced on the insides of the front and
back cover. These two pages, and the corresponding transcript, are reproduced
in Appendix A.
[23] Except for such line breaks that occur in connection with
the phenomena mentioned in the previous paragraph.
[24] It is a common typographic practice to use sic to mark
misspellings and other irregularities which might otherwise be blamed on the
typesetter, and in some contexts punctuation and paragraphing are routinely
normalized, or introduced, by editors.
[25] We note in passing that the statement is silent on what happens in cases (if
any) where the exemplar already contains braces, or an extra space after a
sentence not closed with a period. If there are such cases, then their printed
representation is ambiguous: it could represent the discreet editorial
intervention described in the statement of practice, or it could be a strictly
literal transcript of the paragraph. It seems likely that the editors know
that there are no such cases, and trust the reader to infer it. But it might
also be the case that the editors simply don’t regard such ambiguities as
interesting or important enough to be worth avoiding.
[26] However, this may also be regarded as an application of the
rule of type similarity, cf. our discussion of statement 5.l
below.
[27] If, on the other hand, statement 5.b were taken to mean
that the writers of the manuscripts regarded the use of
capitalization as a matter for individual judgment rather than
orthographic system, resulting in usage that the modern reader
perceives as arbitrary and inconsistent, then the matter would
be similar to the case with u and
v, discussed below. (Readers today do find
the capitalization of seventeenth-century manuscripts
capricious, but we do not believe that is the point being made
in 5.b.)
[28] We take this to mean that the thorn is represented as "th", not
literally as "th," with a comma. Although we do not ourselves follow the
usual rule of placing punctuation before closing
quotation marks rather than after, we are here quoting passages that do
follow this rule.
[29] A glyph shaped like "A" is either uppercase
Latin letter A or uppercase Greek letter Alpha, depending on
context. A dot on the baseline may be a full stop (sentence
punctuation), an abbreviation marker, or a decimal point
depending on both the intratextual and the extratextual context.
In most European languages, "ij",
"ch", and "ll" are each a
two-character string, but the first is a single character (a
single token instantiating a single type) in Dutch, and each of
the others is taken as a single character in traditional Spanish
orthography, even though "i", "j",
"c", "h", and "l" also
occur individually in those languages.
[30] Once again, an alternative analysis might be that the
different type systems are employed for the transcript and the
exemplar. (In this case, the subset relationship would be the
opposite of the one for "u" and "v".) And once again, rules of
default transcriptional implicature would be broken in neither
case.
[32] The transcript appears to be complete with respect to Melville’s notes,
but omits the printed text of Shakespeare which appears in the same
volume.
[33] An alternative interpretation would infer a type system in which for any letter
such as e, there are companion types for
inserted e, canceled e,
overwritten e, and underlined [or italic]
e. Occam’s Razor and convenience in the formalization both lead
us to prefer an account of this example in which insertions, etc. are
compound tokens of corresponding compound types.
[34] The occurrence of a single x is in
fact perhaps strictly speaking ambiguous: statement [5] suggests
the interpretation one undeciphered letter, but
the only occurrence of a single x in the
transcript (in the word extremes on line 37)
seems more plausibly read as transcribing an
x in the exemplar.
[38] Technically, this is obtained by submitting to Vampire the case-specific
facts axiom, a FOF-representation of the axioms 1-8 and of the definitions
a-e, with T-similarity as a conjecture. Vampire’s response is that the
conjecture is a theorem, i.e., that it is logically implied by the other
statements.
[39] Experts on TEI and critical editing will probably find our encoding
idiosyncratic and/or unsatisfactory. Moreover, any serious TEI-based project
today would probably make one base transcription including all the information
contained in both the transcription discussed in [Appendix C] and the
normalized transcription in [Appendix G]. It should be mentioned,
though, that Michael made a more thorough TEI representation of the transcript
for a related project in 2017, using a special TEI-based tag set for
manuscript transcription [http://mlcd.blackmesatech.com/mlcd/2015/W/tip/Melville/Melville-notes.sourcedoc.xml].
[40] This output was slightly retouched by hand to account for a short passage
appearing not at the same place in the exemplar and in the transcript.
[41] It may be plausible to omit unreadable writing from a transcript. However,
in order to leave, for example, text in a different hand or a different language,
an addition or deletion out from a transcript, the transcriber needs to
recognize it as such, and may thus very well be able to read it.
Carter, Clarence E.
Historical editing. Bulletins of the national archives,
Number 7. Washington, DC: National Archives and Records Service, August 1952.
National Archives publication number 53-4.
Grice, H.P.
Logic and Conversation. In Syntax and
Semantics, edited by P. Cole and J. Morgan, vol.3,
Speech Acts. New York: Academic Press, 1975.
Reprinted as chapter 2 of his Studies in the Way of
Words. Cambridge, Mass.: Harvard University Press,
1989, pp. 22–40.
Hayford,
Harrison, Hershel Parker, and G. Thomas Tanselle, ed. Moby Dick, or, The
Whale. Vol. 7 of The Writings of Herman Melville The
Northwestern–Newberry Edition. Evanston [Ill.]: Northwestern University Press;
Chicago: Newberry Library, 1988, rpt. 1994, 1997.
Huitfeldt, Claus,
and C. M. Sperberg-McQueen. What is transcription?Literary & Linguistic Computing 23.3
(2008): 295-310. doi:https://doi.org/10.1093/llc/fqn013.
Marcoux, Yves. A
natural-language approach to modeling: Why is some XML so difficult to
write? Paper given at Extreme Markup Languages®, Montréal, 2006.
Proceedings of Extreme Markup Languages® 2006. On the Web at [https://marcoux.ebsi.umontreal.ca/publi/EML2006/EML2006Marcoux01.pdf], accessed October 2026.
Marcoux, Yves and
Élias Rizkallah. Exploring intertextual semantics: A reflection on
attributes and optionality. Paper given at Extreme Markup Languages®,
Montréal, 2007. Proceedings of Extreme Markup Languages® 2007. On the Web at [https://marcoux.ebsi.umontreal.ca/publi/EML2007/EML2007Marcoux01.pdf], accessed October 2026.
Robinson, Peter, and
Elizabeth Solopova. Guidelines for Transcription of the Manuscripts of The
Wife of Bath’s Prologue. 18 March 2006. On the Web
at zenodo.org, accessed October 2026.
Simons, Gary, Scott O.
Farrar, Brian Fitzsimons, William D. Lewis, D. Terence Langendoen, and Hector
Gonzalez. The semantics of markup: Mapping legacy markup schemas to a common
semantics. In Proceedings of the 4th workshop on NLP and XML
(NLPXML-2004): held in cooperation with ACL- 04 Barcelona, Spain, pp.
25-32. doi:https://doi.org/10.3115/1621066.1621070. On the Web at [http://www.aclweb.org/anthology/W/W04/W04-0604.pdf], accessed October 2026.
Sperberg-McQueen, C. M.
The meaning of OAI 2.0 Markup: An exercise in markup
interpretation. Unpublished fragment, December 2005. On the Web at [http://www.w3.org/2004/04/em-msm/ioai.html], accessed October 2026.
Sperberg-McQueen,
C. M., David Dubin, Claus Huitfeldt, and Allen Renear. Drawing inferences on
the basis of markup. Paper given at Extreme Markup Languages®,
Montréal, 2002. Proceedings of Extreme Markup Languages®
2002. On the Web at
wwp.northeastern.edu (PDF) and www.semanticscholar.org.
Sperberg-McQueen, C. M., Yves Marcoux, and Claus Huitfeldt. Transcriptional
implicature: A contribution to markup semantics.Unpublished, 2018. [XML / HTML], accessed October 2026.
