In an earlier post, I pointed out that the writing of ā word internally in the QCT is very inconsistent, and that sometimes it is written and sometimes it is not. I argued that because words that have internal glottal stop are always written with <ʔ>, unlike many other cases of ā, this might suggest that the internal glottal stop was not lost in these words.
I have been looking into the distribution of written long ā and unwritten long ā in the QCT, and it turns out that there are a lot more patterns than I initially thought, and this puts my conclusions about the internal glottal stop on quite uncertain grounds.
Active participles, which have the shape CāCiC are basically never written with an <ʔ> when they are in the plural with the suffix -ūna/-īna:
kāfir-ū/īna <ʔlkfrw/yn> 'unbelievers', there is a single exception:
It is a little bit messier in the indefinite form, but only for the accusative singular, it is otherwise identical:
The distribution for the construct plural suffixed -ū/-ī, For the construct form CāCiC-ū/ī, Writing with <ʔ> occurs 11 times, while the writing without the <ʔ> occurs 14 times, which is about as close to a 50/50 distribution as you would expect in a corpus.
CāCiC nouns with the feminine plural ending -ātu/i(n) show the same distribution. Words like that occurs 99 time spelled as ṣāliḥātu/i(n) <ṣlḥt> 'good deeds', and once as bāsiqātin <bʔsqt> 'tall'.
For the singular form, which has case ending -a/i/u or with nunation -an/in/un, the picture is similar to that of the construct form (but different to the non-construct plurals), both spellings occur: CāCiC with ʔ occurs 203 times, CāCiC without ʔ occurs 186 times.
Only the indefinite accusative plural seems to have a somewhat significantly skewed distribution, where the spelling without ̣ʔ occurs 59 times, while the spelling with ʔ occurs 33 times. Still some general patterns arise, e.g. wāḥid is never written with <ʔ>, nor is wāsiʕ neither of those two words are actually active participles, but a numeral and an adjective with the same pattern, this might be relevant somehow).
So how do we account for this difference in distribution? Why is 'anything goes' in the singular, while the plural has a very strong distribution? The most obvious solution seems to be taking it as a sign of stress, following the classical Arabic stress rules, the long ā would always be unstressed in the plural (accent the first heavy syllable from the right, excluding the first syllable it encounters, if there are no heavy syllables, accent the leftmost syllable):
The spelling rule thus seems to be that unaccented ā can never be written with <ʔ>, while for an accented ā, it is essentially optional whether it gets written, although there are especially several high frequency words that never get written with <ʔ> (wāḥid 'one', hādihi 'this (f.), ḏālika 'that', wāsiʕ 'wide').
One cannot help but wonder whether this doesn't represent somekind of sound law, perhaps long ā was shortened when unaccented in the language of the QCT, yielding *kafirū́n(a), *kafirī́n(a) and *kafirā́t(u/i(n)).
A similar distribution is found for nouns with the feminine plural suffix -ātu/i(n), which is almost always written with a simple <t>, and not with <ʔt> as is regular in the Classical Orthography, whenever a long ā precedes it in the syllable directly in front of it. In such cases the long ā is never spelled with <ʔ>.
The most common example of this is the noun samāwā́tu/in 'heavens, sky', which is consistently spelled <smwt> 189 times in the Qurʔān (and once as <smwʔt> with the Classical feminine plural suffix spelling, in Q41:12). The 23 other occurrences of this suffix don't spell the preceding long vowel ā either:
ḫālātu/i(n) <ḫlt> 'maternal aunts'
risālātu/i(n) <rslt> 'letters'
maġārā́tu/i(n) <mġrt> 'caves'
ʕalāmā́tu/i(n) <ʕlmt> 'landmarks'
šahādā́tu/i(n) <šhdt> 'testimonies'
ṣāfā́tu/i(n) <ṣft> 'wings spread'
ṣāffā́tu/i(n) <ṣft> 'those lined up'
an-naffāṯā́ti <ʔlnfṯt> 'the blowers'
This is of course very similar to a feature we discussed earlier of Qurʔānic orthography in older documents where words like ʕaḏāb 'punishment' and šāy 'thing', which in the indefinite accusative were spelled as <ʕbdʔ> and <šyʔ>, presumably *ʕaḏabā́ and *šayā́.
The feminine plural suffix spelled simply as <t> (793x) vastly outnumbers the spelling as <ʔt> (a couple of dozen times) in the Qurʔān, and this of course raises the question whether this is not the result of a reduction somehow. The word ḫālāt <xlt> 'maternal aunts' at least precludes the possibility that it was unstressed, as there simply is no syllable left to stress if the ā in the first syllable was also unstressed (unless we place the stress on the case vowel, which would be highly unexpected, considering its ubiquitous loss in all Arabic dialects).
Nouns that write -āt as <ʔt> are a limited set of nouns:
One may consider that within Arabic this word was not considered to have a plural suffix -āt, as it is the plural of bint, with an Old Semitic feminine suffix -t no longer productive in Arabic. Therefore this may have been considered a type of broken plural instead, and therefore avoided the orthographic standard of writing the plural suffix as <t>, alternatively the fact that banāt and bint would become homographs <bnt> sticking to that spelling may have been the motivation.
as-sayyiʔāt <ʔlsyʔt> 'the evil deeds', the ʔ in this word is probably /ʔ/ rather than the long ā.
al-munšaʔātu <ʔlmnšʔt> 'elevated (of sails)', the ʔ in this word is probably /ʔ/ rather than the long ā.
marḍāti (a)llāhi 'approval of Allah', clearly part of religious vocabulary, may have an Archaic spelling.
Already spelled as such in Wetzstein II 1913, carbondated to 662-765 AD.
fī rawḍāti (a)lǧannāti <rwḍʔt ʔlǧnʔt> 'in the lush meadows of the heavens', part of religious vocabulary, may have an archaic spelling. (Q42:22)
Already spelled as such in Wetzstein II 1913.
ṯubātin 'groups', plural of ṯubah unattested in the Qurʔān. No obvious explanation presents itself which this would be spelled with <ʔ> (Q4:71).
Already spelled as such in Wetzstein II 1913.
samāwātin <smwʔt> 'skies', as mentions shows up once like this and as <smwt> 189 other times. (Q41:12).
Spelled as the expected <smwt> in Wetzstein II 1913.
naḥisātin <nḥsʔt> 'misfortune', occurs as such once, no obvious explanation (Q41:16).
Spelled as the expected <smwt> in Wetzstein II 1913.
So of the few words that appear to be spelled with <ʔt> for -āt, two don't show up in the earliest documents, and 3 forms show up early. Two of which clearly are part of religious vocabulary, and it should be noted that outside this one attestation of ǧannāt is always just spelled <ǧnt>.
We can conclude that the spelling <-t> for -āt seems to be a well-established spelling convention in the QCT. It's difficult to decide whether this represents a phonetic reality of the Qurʔānic language, or is simply a spelling convention. The fact that other final -āt sequences are spelled with <ʔt>, at least occasionally, suggests that there was no specific sound law that caused word-final -āt to shift to -at.
Another insight I had on this long ā spelling is that all nouns and verbs with the shape CāC(a) are written with an internal ʔ, regardless of whether they had an etymological glottal stop. The exceptions to this are staggeringly small:
There are 6 examples in total with an unwritten internal ʔ for a CāC stem, consisting of only two lexical items: qāla 'he said', ḥāša, and al-Lāt. qāla 'he said' is actually found much more often in this spelling in older Qurʔān documents (it's basically regular), and this seems to suggest that it was proclitic and unstressed somehow. This is not unusual, quotative verbs "and he said: "x"" are often proclitic (and thus accentless), like the classical Greek φημι (Benjamin Suchard p.c.). That this word was originally proclitic and 'special' in some way in the language of the Qurʔān may be inferred from the fact that some reading traditions require you to not pronounce the final a in qāla, and assimilate the l to following resonants (Ahmad Al-Jallad p.c.).
The other word ḥāša occurs twice in the Qurʔān, in the fixed phrase ḥāša li-llāhi <ḥš llh>, which is supposedly from ḥāšā li-llāhi from the stem III verb of the root ḥšw/y which has the meaning 'to except/exclude' (from s.o.)', Lane explains this phrase as 'I ascribe unto God remoteness from every imperfection or the like, or freedom thereform; generally implying wonder or admiration, so that it may be rendered how far, or how free, s God from every Imperfection!; or this phrase means I seek protection by God; or, as often used by later writers, and in the present day, God forbid!'. Surprisingly, both long ā vowels are not written. I think that this fixed phrase may have simply shortened the vowels in colloquial usage, which spread to the orthography.
al-lāt, one of the Pre-Islamic Arabian goddesses, is spelled as <ʔllt> once in Q53:19, presumably an Archaic spelling.
To conclude: While accented ā seems to have been optionally written with <ʔ> or without, the unaccented etymological ā is never written with <ʔ>. This suggests that it was shortened. It seems doubtful that this is part of the Orthography of the QCT, as such a specific phonetic rule seems difficult to consistently get right if it is not part of the language of the Qurʔān.
A final thought to ruminate over: The 1pl perfective suffix -nā <nʔ> is always written as <n> when it is followed by an object clitic. This suggests that it was unaccented in, e.g. ǧaʕalnā-hu <ǧʕlnh>. This is a little bit surprising if we follow the rules of stress assignment, as this is exactly the environment in which, e.g. kāfirūna has an accented long vowel (vvCv#). It is however also the exact same environment in which the feminine plural suffix is written as <t>. So perhaps there is a sound-law -ā́C# > -áC# at work anyway (assuming word-final vowels were gone).
The other option is that the object clitics did not partake in the stress assignment, in which case the stress would happily fall on the vowel preceding nā.