Issue 334: Scholarly Reading
Posted by Martin on 23/3/2017
I propose to start the discussion about a simplified Inference model for the case in which the interpretation of a text as a proposition is not questioned, but other things are questioned:
A) assertions of historical truth: We need a text with a questioned fact, such as Nero singing in Rome when it was burning. I think Tacitus states he was singing in Rome, and another source says he was on the countryside.
B) Shakespeare's "love is not love" : scholarly interpretation = translation of sense
C) Questioning provenance or authenticity of texts: In the Merchant of Venice, place details are mentioned that only a person who was there could have written that. Shakespeare was not allowed to travel abroad.
C1) Or, critical editions: In the first written version of Buddha's speaches (Pali Canon), there are identifiable passages that present past-Buddha dogmata.
I would start with A), then B), then C)
So, we first want to solve the case that the premise is a proposition, which is not believed as such.
Rather, it is believed that the author of the text meant to express this proposition. This implies that the premise does not make any sense without a provenance assumption, which must be believed.
In A), the provenance of the text from Tacitus is believed. His good will to say the truth about Nero not.
In B) The provenance "Shakespeare" back to the respective edition/name or pseudonym/place of creation is not questioned.
In C1) The text as being that compiled following the first performance is not questioned, but who wrote the text under the name of Shakespeare is questioned.
In C2) The provenance of the Pali Canon edition is not questioned, neither that its content mainly goes historically back to Buddha, but the provenance of a paragraph is questioned.
Therefore, we could Introduce a subclass of I2 Belief i'd call "reading", which puts the focus on believing authenticity of a comprehensible natural language proposition relative to an explicitly stated provenance, but does not mean believing the proposition, nor questioning the intended meaning of the text: see also the ppt
J1 used as premise (was premise for) : IXX Reading
IXX Reading subclass of I2 Belief (or a generalized Belief)
properties of IXX Reading:
JX1 understanding : Information Object (the cited phrase, understanding the words)
JX2 believing provenance : I4 Proposition Set (This contains the link from the cited phrase to the text the phrase is taken from, and all provenance data believed. E.g. Shakespeare edition 1648(??) believed, authorship by Shakespeare questioned, etc.)
JX3 reading as : I4 Proposition Set (the translation of the cited into triples. If absent, the interpretation of the cited phrase is regarded to be obvious)
and J5 defaults to "true" (I believe all "J5 holds to be: I6 Belief Value" should default to "True" if absent).
Then, a conclusion could be that the Information Object (cited phrase) is not believed. In that case, we would need to generalize I4 to be either a Named Graph or an unambiguous text. If we do not, we could use JX1, JX3 to introduce the translation of the cited text as formal proposition, and then use J5 to say "FALSE": "Nero singing in burning Rome 18 to 24 July, 64 AD"
In the case of text sense interpretation, we would need a sort of "has translation" construct, if not simply a work about another work (FRBRoo).
The representation of a text in a formal proposition (Nero P14 performed E7 Activity P2 has type "singing" ...falls within Destruction....)
In the case of the Buddhist text, we would need in addition the believe in the provenance of the post-Buddha dogma, plus the reading, resulting in a different provenance for the paragraph.
If we agree on something like that, let us see if we can simplify or shortcut anything.
Posted by Martin on 30/3/2017
My colleague Athina found the following paper:
Michele Pasin, John Bradley; Factoid-based prosopography and computer ontologies: towards an integrated approach. Lit Linguist Computing 2015; 30 (1): 86-97.
It seems that "factoid" describes the attitude towards a text I tried to formulate as "Reading" ?
Posted by Francesco Beretta on 31/3/2017
I'm quite acquainted with the factoid model and issues it raises, and participated recently in a workhop about prosopography in Vienna (in February) where I met once again John Bradley. I would prefer to discuss all this in Heraklion or at a later point on the list because the issue is quite complicated and needs a thorough discussion.
Juste to give some examples quoting the Bradley/Pasin text:
"Factoid represents the spot in a primary source where something is said about one or more persons".
"factoid approach prioritizes the sources, rather than our historians’ reading of them. "
"state of affairs described by the document is either a ‘situation’ or an ‘event’, i.e., more generally, a ‘temporal entity’"
As you see, there are many notions behind the factoid model and I couldn't say for sure if it's an event, a reading or a portion of text, or all of them at once.
In the same paper Martin mentions, an alignment with the CRM is proposed: a factoid is a subclass of E2 Temporal entity and a so called "document-interpretation-act" is a subclass of E3 Attribute assignement. The reading would be more on this side.
I'm not (yet) so trained in the CRM to be able to tell if this alignement is accurate but in any case this paper raises very interesting issues we have to discuss (and are discussing) in the domain of an extension of the CRM for historical data. And are certainly worth a discussion in the SIG.
Posted by Francesco Beretta on 4/4/2017
Here some interesting documentation about the Factoid model:
posted by Simon Spero on 5/4/2017
A quick meta-point on the issue, and the term factoid.
Posted by Anaïs Guillem on 5/4/2017
following the discussion during the sig meeting, here is the link to the diagrams representing the conversation with the Nero’s example that can be used for HW:
Feel free to comment
Posted by Martin on 5/4/2017
Dear Simon, All,
Your comments well taken, just to clarify what CRM Inf intends to do:
It does never aim at replacing scholarly and scientific inferencing by an automated machine. Its sole purpose is to document the steps of scholarly inferencing, provenance and dependencies of knowledge used and created. Any use of logic to automate the transition from premises to conclusions would be documented as a use of a tool. In particular, it does not aim at simplifying the scholarly process, only at explicating assumptions and intermediate steps to the degree they are an apparent stage at which other actors would take up results and continue. The CRM is not an AI endeavor.
Personally, I do not regard that scholarly inferencing can comprehensively be described in an adequate manner by any formal logic, at least for the next years to come.
What I have proposed, does not simplify inference. It just describes a frequent case, in which the propositional interpretation of a piece of text in regarded as unambiguous, but the truth of the propositional content is subject to an inference. The construct I propose just models this premise. It assumses, that necessarily any scholar making such an inference, must have an explicit believe about the provenance and authenticity of this piece of text, and that his conclusions will depend on the truth of this provenance assumption.
CRMInf so far had no construct to describe such a premise in a comapct way.
I did not make any statement about how the provenance assumption has been achieved, nor what its precise form is, nor that scholarly reading does not imply questioning the propositional interpretation, which would be just a case just not modelled here. Nor did I make any statement about the following kind of inference.
The simplification is only in the complexity of the model for cases in which defaults are known to hold, here, that the propositional content of the text has not been questioned by the actor making the inference. It is a sort of shortcut.
But the same construct can also be used as a conclusion about the propositional content of a piece of text.
We can never, in a realistic information system, be explicit about all steps of argumentation. We have to rely heavily of defaults. When starting reading a manuscript or inscription, a process of endless details starts until we come to sets of alternative propositional interpretations. Most of these cases can only effectively be described in form of explications to otherwise implicit arguments. The art will be to find out which series of arguments can reliably be documented in an implicit way, and how later details can be added without creating a non-monotonic representation.
If anybody still regards this as an attempt to simplify inferences, I'd like to be pointed to my errors