Editing Openai/69186d20-2d14-8013-b9c6-c09000f173f6 (section)

==== Timeline, very compressed: ====
# Proto-Celtic / early Old Irish - Had real geminate consonants: /ll, nn, mm, rr/ with extra duration. - These came from: - inherited IE geminates, - assimilation across morpheme boundaries, - analogical formations.
# Old Irish - Geminate vs single distinction in sonorants gets reinterpreted: - Geminate ll, nn, mm, rr → “fortis” sonorants (stronger, tenser) - Single l, n, m, r → “lenis” (weaker) - At this stage, length is no longer the main cue; you get a bundled cluster of phonetic effects (duration, articulation, maybe lowering of preceding vowel). - Crucially: this fortis/lenis contrast is now phonemic in some positions.
# Middle → Modern Irish - The functional load of fortis vs lenis shrinks: - Many former contrasts are lost or neutralized in common environments. - Broad vs slender becomes the salient contrast (and is reinforced by spelling). - Orthography keeps ll, nn, mm, rr in lots of places, but now the fortis realization is mostly reduced to subtle timing/quality differences, not a clear-cut second parameter.

So to your specific options:

: 

Best description:
* There was true gemination originally.
* The geminate vs non-geminate contrast then turned into a fortis vs lenis contrast (same segment quality axis, different primary cue: tensity rather than raw length).
* That fortis/lenis contrast has since been mostly absorbed into: - broad vs slender, - general timing/strength patterns, - and lexicalized idiosyncrasies.

So: fortis sonorants are historically “ex-geminates”. They’re not some independent ancient tense/lenis system that gemination later merged into; rather, gemination was reanalysed as “fortis”, and then the phonetic salience of this fortisness has faded over time.