We don't all agree what a solution needs to address.
Various kinds of "authors"
Pure form of single authoring is considered impractical except for the lowest common denominator, which fails to meet the content providers needs wrt end-users
Not going to be solved by one language.
Application consists of n web pages and need to target m devices, giving rise to a n times m problem. Exhaustive approach doesn't scale.
It should be easy as possible to write and maintain content for multiple modalities/devices.
Authors must be able to control the presentation for specific modalities/devices. But this not required in all applications.
It shouldn't be more work to use this solutions/tool than writing separate pages for each modality/device
Some authors don't care about modalities, and the authoring system should be able to fill in for the roles of the graphic designer etc. for the different modalities.
Abstract models/device classes help to reduce the complexity in tailoring content.
A transformational solution involves a separate transform (style sheet or script) that maps content into a form appropriate to a specific set of modalities/devices.
The semantics need to be present in the raw content in order for a transformational approach to be able to tailor the content to best suit particular modalities/device classes.
Application model has to address state, interaction and transactions. Not all applications will use all of these.
Another approach is to embed information about presentation and interaction into the content. Hints and graceful degradation models are used to tailor to particular modalities/devices.
Constraint-based mechanisms can simplify the task of the author by pushing the burden back on the implementors.
Evolution from experience in hand-tailored customization leads to recognition of design patterns that facilitate re-use, which then enable the construction of authoring tools that reduce the effort.
Mass market solutions versus high-end high-volume solutions.
How much has to be standardized?