Latin liquid dissimilation and the relevance of phonological structure

The Latin suffix -alis surfaces as -aris after a stem [l] unless [r], a dorsal or the glide [w] intervenes, blockers McMullin takes to be no natural class. Jacobs puts [lateral] under a LINGUAL node beside CORONAL and DORSAL, specifies it only on liquids and leaves CORONAL unspecified, so that the segments bearing LINGUAL are the liquids, dorsals and glides. Dissimilation on that tier derives the forms the paper accounts for and misses the rest, where the tiers of Steriade, McMullin, Cser and Belth each miss a form. Jacobs's constraints select the rule's output, and a [j] from a vowel in hiatus blocks dissimilation.

Implementation notes

• Segments are the Fragment's, with PHOIBLE's values. underlying reads Jacobs's representations off them: [lateral] only on liquids, after Steriade, CORONAL never, and LABIAL and DORSAL where present. The tree is Jacobs's statement of the LINGUAL node, which he attributes to Clements and Hume; the attribution is his. The tier is that of C-place, so a segment on it is [−syllabic]: a vowel in hiatus is off the tier, the glide on it.
• The rule is a SearchCopy whose target is the suffix liquid L, unspecified for [lateral]. Jacobs's input has the suffix -alis, which the tableau of his constraints takes. Only the suffix liquid alternates, as his ban on adjacent laterals is restricted to the suffix and Latin has stem-internal [l]…[l].
• A row is compared on the paper's underlying stem, such as latj for Latialis and glajb for glebalis, with the suffix liquid read off the attested form.
• Steriade's tier is derived like Jacobs's, as the segments specified for [lateral]. McMullin's tier and Cser's blockers are listed as their sources list them, and Belth's rule is his study's, over its own transcription. Roberts's correspondence account is known here only from Jacobs's report and is not formalized.
• McMullin's grammar bans tier-adjacent [l]…[l] over whole words, which also rules out his own stellaris, whose geminate puts [l l] on the tier. The rule here changes only the suffix liquid.
• dissimilate_glide applies to every stem whose last LINGUAL segment is [l], so it also predicts later -alis forms of tricliniaris and of Cser's Saliaris, which the paper does not check.

Errata

• The paper predicts that the -alis forms of the hiatus pairs "did not occur prior to 200AD", but its own table of counts has Latialis three times before 200 AD.
• The variation is said to occur "only … with intervening coronal consonants" and a front vowel in hiatus, but fulminalis and fulminaris both occur in the 5th century, with [m] and no hiatus, and the rule derives only fulminaris.
• [j] is described as [CORONAL] and [DORSAL], as "lingual [dorsal]", and in a diagram as [lingual] over [coronal]. With CORONAL never specified, the diagram's [j] is off the tier and could not block; the account needs [DORSAL], as in PHOIBLE.
• One representation is captioned pluvialis but spells fluvialis, the list of dorsal blockers is cited under the wrong example number, and Cser's counts for Latialis are cited a page late.

TODO

• Cser's further condition, at least three moræ between the stem and the suffix liquid, needs vowel length, which the Fragment's segments do not carry. limitaris is the rival case either way: its [m] blocks for Cser, and it has more than three moræ.

References

• Jacobs 2026, Latin liquid dissimilation and the relevance of phonological structure
• Steriade 1987, Redundant values
• McMullin 2016, Tier-based locality in long-distance phonotactics: Learnability and typology formalized
• Cser 2010, The -alis/-aris allomorphy revisited
• Cser 2020, The Phonology of Classical Latin
• Belth 2026, A Learning-Based Account of Phonological Tiers formalized
• Clements & Hume 1995, The Internal Organization of Speech Sounds