Robert Forkel

> Not allomorphs.csv, since it links allomorphs (FormTable) to morphemes (MorphemeTable)? The allomorphs would be the things in `FormTable->Form`. And they'd be grouped into morphemes via a `morphemeReference`, pointing to...

I'd put `morphemeReference` directly in the `FormTable`. In the simplest case, this wouldn't even have to link anywhere, but just be a value to group by, like Cognateset_ID for cognates...

@cysouw `Segment_Slice` is separator agnostic. It applies conceptually to the CLDF data "as read", i.e. after the separator has been used to convert a string on disk to a list...

@cysouw If you say `Form` was chosen to avoid the wordhood problem, wouldn't it be transparent enough to have morphemes in `FormTable`, but add a `morphemeReference` column - thus marking...

@cysouw I think in absence of a separator, slice notation does not make sense. We could interpret strings as lists of characters - but that wouldn't be well-specified - in...

@cysouw I don't fully understand. Do you mean to say that you interpret `FormTable` differently than how it is typically used - in the lexical datasets that are out there?...

I don't really understand. So do you mean > if a table has an #id column, it must be declared as primaryKey and if it declares a primaryKey, it must...

Regarding the second point: `pycldf` does that - sort of: It adds foreign key constraints to the metadata when you put together the metadata object programmatically, see https://github.com/cldf/pycldf/blob/273f424e03401ceecf0fe07dac08793a2602b254/src/pycldf/dataset.py#L344-L349 So when...

I would not want the spec to dictate what to use as primary key of a table. It's clear - I hope - that any #id column must be unique,...

I see that more constraints would make tooling simpler. But there's also the (somewhat secondary) design goal > it is often possible to retrofit existing CSV files into the CLDF...