a post to coordinate thoughts on a mastodon conversation with @[email protected]

introduction

@themanual

https://mastodon.online/@themanual4am/111997350490358955

…I found a deep mapping between the accessibility of science writing and software code, based around the idea of domain translation (see images).

Though, I feel the accessibility issues you refer to here are more ’expressive style/ adornment'?

Is that right? Can anyone provide examples?

1/

@themanual

link

Software also includes tools to programmatically detect and remap styles and other presentational characteristics. This helps to onboard individuals to specific standards. And also retrospectively, somewhat ’re-style or re-present’ legacy or differently presented documents, to modern taste.

I guess I’m curious to see how/ whether the two relate.

Many thanks.

2/

@axoaxonic

link

  1. The thing I wonder about with this is in possible cases where the style encodes the meaning in an immutable way, that problem of hyper-specificity being used in scientific writing to describe things that would be really difficult to described otherwise. Maybe there could be a way to attach clear, general descriptions of those concepts that can’t be clarified through style transfer

  2. Kind of repeating things that have already been said, but approaching it from a style or translation might always run into cases where certain terms and phrases are “hard coded” into their contexts, untranslatable without losing something.

  3. One way of approaching this could be looking at underlying meanings, not the symbolic referent but their foundational categorization or representation. Not the type of rule based translation translation, via term → equivalent term, but translation methods that have an intermediate layer for meaning, like term → meaning → equivalent meaning → [any way of expressing it]. I think it would help to see which concepts could be mapped at all and to where. Automated analysis of meaning looks difficult, I only know of Lambek Calculus and Typelogical Grammar but am not very deep into computational linguistics, but a quick glance at en.wikipedia.org/wiki/Machine_… shows some examples of translation methods that have the meaning layer

  4. Just wanted to add this because equivalent programs with different styles could be judged as equivalent by their outputs, but things are subtler with language I think, because of the meaning/concept/representation aspect of cognition.

response

This is great @[email protected], thanks (apologies again for the extended delay!).

Firstly (summarising): you’re confirming that style ‘is more than presentation’; that style occasionally encodes concepts (as if a domain-specific shorthand? That work?)

If we consider style as a secondary channel of communication (with explicit linguistic-conceptual map/prose the primary), and in the context of the broader conversation (#earlier @axoaxonic), are the ‘weird old traditions’ you refer to exclusively the second channel, or also first? And are you able to provide any examples?

point by point

a breakdown of my interpretation of the above four points

  1. A situation where style communicates something which is challenging to describe in terms/ phrases (style as disciplinary shorthand; a second channel; either discreet phenomena or primary channel modifier?)

style communicates additional special-domain concepts through a second channel which must be made explicit to capture all detail

  1. A situation where a special-domain concept (corresponded to by some special-domain term/ phrase) is inseparable from broader special-domain context, which complicates decontextualisation of concept from special to general-domain

i think complicates more than prevents; i don’t believe there is such a thing as a completely isolated special-domain concept which cannot be reframed in ‘specially recontextualised generalised concepts’, as this implies *’territory by magic’

  1. So, you’re making a distinction between symbolic referent and representation – can you say more? This sounds like intermediate generalisations (above) – like the difference between partial/ embedding and a whole?/ some sensory pattern-fragment and the corresponding derived accumulation?/ or something else?

  2. Follows from 3, but a situation of {false positive heuristic; shallow equivalence}

programs which result from different code-bases might produce circumstantially equivalent output

style

So, I think that you refer to the tooling/ programmatic side of style translation specifically.

and for me, tooling follows understanding: not this other – ‘just ML it and use whatever pops out in production’ malarkey

And yes, in cases where style encodes concepts, inexplicit presentational translation would be insufficient.

To qualify, I think that the term ‘style’ applies to plural phenomena :

  1. Encoding (linguistic-conceptual)
  2. Expression (explicit linguistic sequence, implicit conceptual composition)
  3. Presentational (adornment), at minimum

Note, (1, 2) are not covered by the above style tooling example.

I singled out style (above) because, for language, I considered style adornment (3) to be a distinct channel of communication (secondary to the underlying conceptual synchronisation of 1, 2); which can in-fact also communicate high-dimensional concepts in a ‘special domain shorthand’ kinda way; and so adornment translation likely also needed to be conceptually mapped. Without examples I wasn’t sure, but you seem to qualify.

  • Had no visibility

headlines

  1. Software code is more accessible than science writing because software has a well-defined general-domain, to standardise special-domain composition; and science does not (but ought to)

inaccessibility is one aspect of technical-debt. refactoring technical-debt, by principled iterative recomposition of special-domain concepts, around standardised principles, primitives and intermediates, of a well-defined general-domain, improves accessibility, productivity and quality.

    • My analysis
      • Paraphrasing, I consider terms/ phrases as being ’effectively arbitrary externalised pointers’ to underlying conceptual/ representational form (which captures body-state, etc by proxy)
    • Implicit is that
      • Behind every linguistic accessibility edit is a corresponding conceptual form (/ refactor); in-fact operationally, that conceptual form must exist or that refactor must have occurred, for the accessible term to make sense
      • Is based upon the concepts behind language
      • To make special-domains accessible, we must recompose concepts from more fundamental forms (with respective corresponding terms)
    • Refactoring as distinct from variable renaming
      • There is something that it is to rename variables to better correspond to characteristic profile
      • There is something that it is to restructure characteristics to
    • Generally I consider that
      • Everything communicated by any channel {terms/ phrases; style; other} resolves to concepts (albeit across arbitrarily plural contexts, eg: conceptual domain; human in-group domain)
      • Accessibility is an exercise in making the implicit explicit: if education is isolated holistic conceptual onboarding, accessibility is integrated relative conceptual onboarding
      • The well-formed general-domain above is a way to standardise conceptual onboarding across all domains (holistic and integrated), because (i suggest) this reflects the nature of our continuous territory
    • I ought to have framed the style question better
      • I’m pointing at your observation of {weird old traditions; technical pseudo-regal style} and wondering whether
        • This refers to purely cosmetic adornment (in-group signalling, etc); or whether I need deeper contextual/ contextual handling; is the inaccessibility of it all a result of cost-of-parsing overhead (like an interpretive handicap) more than insufficiently-explicit conceptual definition;

state, staging, validation

  • Software code repositories are like a live-document; and software explicitly allows (and at times encourages) staged progression towards ideal state
  • Clean (well-formed) code takes effort and time; so (for non-trivial projects) when the value in development objective is yet to be proven, software encourages ‘minimal-viable product’ (or prototype) to prove the objective (in some manner, for example problem-solution fit demonstrates that a problem has been solved, in principle)
  • After this, the software can be re-engineered for production/ release (device performance, infrastructure scale, etc) – this is really no different to a physical ‘garden-shed’ prototype demonstrating a capability, before being re-implemented for mass-production by a factory production line, etc
  • Some products are ‘ahead of their time’, in that we can demonstrate a prototype, but lack the means to progress implementation to release form
    • We lack the knowledge to cleanly re-implement the redundant complexity of a prototype to an optimal form
    • Once we identify and solve that separate problem, we then re-implement the prototype around its ‘conceptual/ practical essence’

notes

  • Thinking of science writing accessibility in terms of code makes a very clear case for accepting ‘accessibility staging’
  • How does staging relate?
    • Some scientific ideas resemble production/ release; others resemble prototypes (a literal proof-of-concept, of course). The proof-of-concept demonstrates that a problem can be solved, though is not yet in release form. In science and software (though more enterprise I suppose), people use proof-of-concepts in production all the time
    • I think that science sometimes forgets that ‘form isn’t finalised simply because use is found’, that useful form is not meant to be the final form
    • Too many prototypes become sentimentally revered; however useful, many problems which exist with a prototype form disappear when state is refined and progressed
  • Refactoring code is an integral part of code-lifecycle; debugging code is more difficult than writing code; more time is spent refactoring and debugging code than the initial write
  • Subsequently, code ought to be written to be read, by humans as well as compiler
    • There is a difference between clarity and expression brevity; brevity which increases interpretive ambiguity

appendix

earlier @axoaxonic

link

I’m hyperlexic and a lot of the experiences I have with insufficient speech are because I can’t find ways to express my thoughts in ways that wouldn’t confuse the neurotypical people. But I want to be able to communicate with anyone, so I translate or mask as much as I can, but it’s hard to do that in real time. Often I just say nothing or give a minimal response.

With writing, if I want to reach a wider audience with diverse neuro types and styles of parsing info, I can edit, listen to my own drafts with a TTS program etc. It’s still very difficult to even imagine how someone’s going to parse my thoughts though, let alone figure out how to accurately package those thoughts in accessible language. Writing classes can teach nice ways of writing, but overcoming the “double empathy problem” in sci com would be a whole different challenge.

I think the majority of inaccessible academic writing, though, comes from weird old traditions in academia, lots of authors emulate and repeat aspects of the same technical pseudo-regal style that’s been going on for centuries.