Abstract: There are various proposals for how to represent contrastive stress in the lexicon, and many rely on underlying forms that are already grouped into prosodic units like feet and syllables. I will argue that such theories cannot account for the stress system in the Northwest Caucasian language Abkhaz. Similar to some Indo-European languages (Greek, Russian, … ), Abkhaz has multiple accent specifications per word, which then together determine the placement of surface stress in a predictable way.
Using data from an Abkhaz dictionary, I analyze 600+ stress alternations, and phonotactic restrictions across 6,000+ native and borrowed words, the largest dataset on Abkhaz stress to date. I show that the number and type of stress alternations are straightforwardly explained by a distinctive feature [stress] attached to individual segments. Other proposals, where stress is a property of morphemes (Dybo 1977), syllables (Trigo 1992), or moras (Vaux and Samuels 2018), all make incorrect predictions. This shows that in at least some languages, stress is a segmental property in the lexicon.