Re: arXiv updates and MathML Intent

It's likely "box" is good enough, but as you hint at, not if there is a
white box and a black box in the paper. It would require seeing the whole
document to know whether there is one or more look-alikes in a document. I
can't see that happening without AI being the one that figures out the
speech after looking at the document. On the other hand, the author surely
knows that info and more importantly, knows how they speak about it when
they talk to colleagues. Intent solves that problem. The real problem is
how do we make it trivial/a part of the author's workflow to convey that
information.

     Neil


On Mon, Sep 22, 2025 at 12:35 PM Paul Libbrecht <paul@hoplahup.net> wrote:

> Hello Deyan,
>
> What the spec does not (and probably cannot) say is how much
> “decontextualised” it should be spoken.
>
> If all sorts of other squares or boxes were there, then pronouncing “white
> square” or “empty square with thin border” (?) may be useful but if, as in
> this case, it is the only box, then pronouncing “box” is likely enough.
>
> I interpret, personally, literal very close to a decontextualised version…
> But it might not prevent to say box… as long as “the literal expectation”
> would not be hurt.
> This is probably space for experiments and research and for later versions
> of MathML.
>
> Keep up the inspiring examples!
>
> Paul
>
> On 22 Sep 2025, at 21:15, Deyan Ginev wrote:
>
> Hi Paul,
>
> The main reference for :literal is likely at:
>
> https://w3c.github.io/mathml-docs/literal-common-properties#example-defaults
>
> To my understanding, ATs have freedom how to vocalize the literal
> character content of U+25A1 □
> In particular, an AT may want to localize the character to a desired
> natural language different from English.
> Speaking the Unicode name is certainly a very reliable starting point (it
> is already well-defined on all inputs).
>
> But if there is a better way to narrate than "white square" and an AT
> implementer wants to provide a consistent experience where they don't
> tightly follow all Unicode names, that should be fine as well I think.
> The key directive prescribed by :literal is to describe the written
> contents, so I expect possible alternatives to be akin to "hollow square",
> "square frame symbol", etc.
>
> I expect for a while all implementers of :literal will rely on the Unicode
> names, but the future often surprises me.
>
> Greetings,
> Deyan
>
>
> On Thu, Sep 18, 2025 at 1:36 PM Paul Libbrecht <paul@hoplahup.net> wrote:
>
>> Hello Deyan,
>>
>> So what is supposed to be the pronunciation of the box symbol there
>> supposing to represent the cartesian product of graphs ? □
>>
>> According to:
>> https://w3c.github.io/mathml-docs/unicode-speech/
>> It would be called (for 25A1 □): *white square*
>> Similarly, the double-lined-right-arrow ⇒ should be called *implies* but
>> would be called “rightwards double arrow”.
>>
>> Am I understanding it right that this is what literal would mean? Is this
>> “left to interpretation to the AT” ?
>>
>> Paul
>>
>> On 17 Sep 2025, at 23:39, Deyan Ginev wrote:
>>
>> Hi everyone,
>>
>> As of this morning, arXiv's experimental HTML pages are now using an
>> intent attribute for the first  time. Namely, each math element is
>> annotated with the ":literal" property.
>>
>> ":literal" values are the only reliable starting point for graduate level
>> texts in STEM, where simple heuristics fail as often as they succeed (if
>> that). They fit Louis' pragmatic stance of "just show me what the author
>> has written", while also allowing more informed intent attributes to be
>> deposited on the subtrees, when known.
>>
>> Additional intent upgrades are interesting future work for arXiv - and
>> LaTeXML. We would like to gradually enrich subtrees, whenever we have
>> sufficient confidence we can infer the mathematical concepts.
>>
>> These new attributes for arXiv are only starting to trickle in with the
>> article batches announced daily. It will take a full regeneration of the
>> collection to have them everywhere. That could happen near the end of 2025,
>> maybe just in time for the MathML 4 CR phase. We'll see.
>>
>> Here is one random recently announced article, to illustrate today's
>> changes:
>> https://arxiv.org/html/2509.12354v1
>>
>> P.S. My regrets for tomorrow's WG meeting, I will be traveling.
>>
>> Greetings,
>> Deyan
>>
>>