Alex & I have been recently talking about how to do "evidential-type
things" in our gripping language.

I'll quote his writeup from our design doc. PF = 'pronominal frame'
and has to do with logophoric shifting of deixis in subclauses;
"Ascriptor" is a word that Alex just made up, AFAWK.


The ascriptors are words used to frame beliefs, assertions, and
judgements of other people, sources, or standards, and possibly to
describe justifications for these.  Ascriptors have a certain amount
in common with evidentials (well, if evidentials took source
arguments), but they fundamentally differ in that saying something
with an ascriptor makes no inherent claim to its truth (like good
little skeptics...), whereas using an evidential does -- "Bob asserts
X" vs. "X is true, and I know 'cause Bob said so".   Perhaps they're
more akin to verbs of belief.

The argument of an ascriptor specifies the source of the information.
This argument often defaults to the effective speaker, where that is
sensible.  From this perspective the most unmarked one is [claimed-by]
which adds no semantics beyond asserting a belief: in particular when
used intransitively it's a no-op.  They can nest: [claimed-by] [Bob]
[experimental-evidence] [Fletcher] <claim> "Bob reports an experiment
of Fletcher showing (claim)".   Ascriptors used with an argument other
than the current effective speaker trigger a pronominal frame shift,
as in the last section; in particular if there is a shift ascriptors
nested inside that one default to the effective speaker of the
subordinate pronominal frame.

Adding a [claimed-by] where it would be possible to do without, e.g.
[claimed-by] [Bob] [experimental-evidence] rather than
[experimental-evidence] [Bob], adds an implicature of skepticism.

Using [claimed-by] with a predicate with subjective value gives it a
sense more like "as deemed by" or "in the opinion of".  This for
example would be used to render the sense "like": "Bob likes X" would
be [claimed-by] [Bob] <X> [good].  Other words with this flavour of
resolving subjectivity include [by-rule], which can take a ruleset
argument, and refers to an unspecified ruleset (rather than the
effective speaker) without one.  (Does it really make sense for these
guys to shift PF?)"

I have two questions.

First: Is it true, as Alex asserts, that in natural languages
evidential uses *always* contain a connotation of assertion by the
speaker, rather than mere report? Are there natlangs where the classes
of 'evidentials' and 'reported speech grammar' (whatever the latter
happens to be) difficult to distinguish?

Second: What do y'all think of the Sapir-Whorf style interpretation of
this as a cultural thing?

Very crudely put:

a) no evidentials: people will tend to incorporate beliefs of others
as if they were their own, and assert them as their own
b) assertive evidentials: people will do the same, but be more likely
to report them with caveats (e.g., if they heard it by rumor, they say
so and thus do not make a strong assertion that the former speakers
c) nonassertive evidentials / 'ascriptors' (as above): people will be
less likely to incorporate others' beliefs as their own, but instead
carry with them a bunch of claims-by-others that they might or might
not have verified themselves

It seems to me that, if SW is true at least moderately, a) will lead
to more sheephood and c) to more skepticism.

- Sai