The role of heads and cyclicity in bracketing paradoxes in Armenian compounds

Abstract

It is often argued that words have complex internal structure in terms of their morphology, phonology, and semantics. On the surface, Armenian compounds present a bracketing paradox between their morphological and phonological structure. I argue that this bracketing paradox simultaneously references endocentricity, strata, and prosody. I use Armenian as a case study to argue for the use of cyclic approaches to bracketing paradoxes over the more common counter-cyclic approaches. I analyze the bracketing paradox using cyclic Head-Operations (Hoeksema 1985) and Prosodic Phonology (Nespor and Vogel 1986), specifically the Prosodic Stem (Downing 1999a). I argue that the interaction between the bracketing paradox and the rest of compound phonology requires the use of stratal levels and cyclicity. I argue that counter-cyclic approaches like Morphological Merger (Marantz 1988) or Morphological Rebracketing (Sproat 1985) are inadequate because they make incorrect predictions about Armenian phonology.

Appendix: Counter-cyclicity with alternatives to stratal phonology

Setting aside prosodic variation, I showed that counter-cyclic approaches like Morphological Merger are partially adequate for modeling the bracketing paradox but they contradict the stratal phonology of Armenian. I instead argued that the bracketing paradox required a cyclic approach like Head-Operations which did not contradict the stratal phonology. In this appendix, I sketch out two alternative analyses that combine counter-cyclic approaches with two alternative to stratal phonology: phase-based phonology and Halle and Vergnaud (1987a)’s free interleaving of cyclic and non-cyclic phonology. I argue that these alternatives have problems in modelling the entirety of compound phonology and morphology.

1.1 A.1 Empirical problems with phase-based phonology

Phonological Derivation by Phase (PDb) is a conceptual offshoot of lexical phonology (Marvin 2002; Newell 2008; Samuels 2011, 2012). While several theoretical assumptions of PDbP are still being debated (Matushansky and Marantz 2013; Siddiqi and Harley 2016; Newell et al. 2017), the following discussion refers to the more commonly held assumptions. I argue that when combined with Morphological Merger, PDbP can model the bracketing paradox and the application of stem-level rules in endocentric compounds. But it has problems in the application of stem-level rules in exocentric compounds.

PDbP assumes that words are cyclically derived where cycles are defined in terms of phase head n, v, a (= derivational suffixes). Non-phase heads (= inflectional suffixes) cannot trigger spell-out and thus do not trigger any cycles. They are not just post-cyclic, but they are part of the same cycle as derivation (Embick 2010). PDbP generally assumes that there is only one cophonology (= no stem vs. word strata) and that there are no prosodic constituents (Scheer 2011, 2012). Cyclic derivations generally respect the phonological structure created on previous cycles, whether by the Phase Impenetrability Condition (Marvin 2002), Phonological Persistence (Newell and Piggott 2014), phase-based faithfulness (McPherson and Hayes 2016), or the use of cophonologies (Sande et al. 2020).

Recall that destressed high vowel reduction (DHR) applies in derivation, but not inflection. It applies in both endocentric and exocentric compounds.

• (54)

  

First, without strata, PDbP cannot directly capture the fact that inflection can trigger some but not all stem-level processes. The null hypothesis is that inflection cannot trigger its own cycle because inflection is a non-phase head. Instead, inflection is incorporated into the same phase-cycle as the previous (covert or overt) derivational suffix (Embick 2010). For simplex stems, this would allow inflection to trigger stress shift but not DHR.

• (55)

  

When combined with Morphological Merger, PDbP can model the bracketing paradox. Reduction correctly applies in endocentric compounds; howeveer, reduction incorrectly applies in exocentric compounds. I assume the following morphological structure for endocentric and exocentric compounds. These structures are adapted from conventions in Distributed Morphology (Harley 2009; Harðarson 2016, 2017, 2018). Each stem is made up of a root and covert category affix. In the endocentric compound, Stem1 is adjoined to Stem2, as indicated by the different indexes for n. In exocentric compounds, Stem1 is also adjoined to Stem2. The entire compound takes a covert adjectivizer a (Steddy 2019). As before, I assume that linking vowels are generated in phonological spell-out in the PF component as dissociated morphemes (Embick 2015).

• (56)

  

I go through a derivation below. Following morphological rebracketing, the pl forms a constituent with Stem2 in endocentric compounds. The pl is spelled-out as -er in the first cycle with Stem2. The two stems are concatenated in Cycle 2 and reduction correction applies.

• (57)

  

But, reduction is incorrectly predicted to apply to Stem2 in exocentric compounds. In the first two cycles, each stem is spelled-out and then concatenated. By the second cycle, the compound has undergone stress shift to Stem2 and reduction on Stem1: . At this stage, the singular compound is an endocentric noun, but it must be reinterpreted as an exocentric possessive by taking a covert adjectivizer a (Steddy 2019). But this causes a problem for the phonology. In Cycle 3, the a category-node for the compound is spelled-out alongside the pl. Although the correct plural is generated, the monostratal phase-based phonology must apply stress shift to pl and reduction on Stem2: *.

An additional conceptual problem is that the above PDbP system contradicts Phonological Persistence. Phonological Persistence (Newell and Piggott 2014) predicts that structure-changing processes (reduction) should be possible within the same phase. By being in the same phase-cycle as the root, inflection would not be blocked from changing any structure: reduction would be incorrectly predicted to be preferred. In contrast, derivation would be incorrectly predicted to only trigger structure-building processes (no reduction) because it is in a separate phase.

In order to get the right reduction patterns, the simplest solution is to allow inflection to trigger its own cycles (Bobaljik and Wurmbrand 2013).²⁷ Phonological Persistence would then block inflection from triggering DHR. However, this implies an informal stratification of phonological processes. Furthermore, we still need a separate mechanism that allows derivation to trigger structure-changing processes like reduction.

To handle Armenian, PDbP can be tweaked to include strata via final vs. non-final phases (Lochbihler 2017), separation of phonological and morphosyntactic cycles (Embick 2014; d’Alessandro and Scheer 2015), and allowing inflection to trigger its own cycles (Bobaljik and Wurmbrand 2013; Shwayder 2015; Kilborne-Ceron et al. 2016). Doing so reduces the differences between PDbP vs. Stratal/Lexical Phonology (cf. Bonet et al. 2019). This brings us back into the original problem of how Morphological Merger contradicts stratal phonology.

1.2 A.2 Conceptual problems in the free interleaving of cyclic and non-cyclic phonology

Phase-based phonology assumes that it is morphosyntactically predictable if a given morpheme will trigger a cycle or not. In contrast, the model of phonological cyclicity in Halle and Vergnaud (1987a,b) removes this predictability. Here, where an affix can trigger cyclic rules or not is diacritically determined. There are also no constraints on the ordering of cyclic and non-cyclic affixes. Scheer (2011, 9) calls this system “selective spell-out”. Furthermore, PDbP assumes a single cophonology or set of rules. In contrast, Halle and Vergnaud (1987a,b) assume two cophonologies: cyclic vs. noncyclic rules. Whether a phonological process is cyclic or not is diacritically determined. I show that this lack of constraints makes it possible for Halle and Vergnaud (1987a) to model the Armenian data. However, this model does not find independent evidence elsewhere in Armenian.²⁸

Consider again the application of destressed high vowel reduction in non-compounds. To make the comparison easier, I decompose stress shift into two processes: rightmost stress assignment and destressing. Informally, Rightmost Stress Assignment places stress on the rightmost full vowel; Destressing removes stress from a stressed vowel which is not the rightmost stressed vowel.

In my analysis, reduction is a stem-level rule which applies in derivation (the MStem) but not inflection (the MWord). Stress-assignment and destressing are both stem-level and word-level rules. In Halle and Vergnaud (1987a,b)’s system, stem-level rules are called cyclic rules while word-level rules are noncyclic. All derivational and inflectional suffixes are diacritically marked as triggering cyclic (_c) and noncyclic (_nc) rules respectively. The concept of MStems vs. MWords is not relevant.

• (58)

  

To match the representations used in Halle and Vergnaud (1987a), I do not assume covert category suffixes on roots. In Cycle 1, the noun is spelled-out. Because the noun has a cyclic diacritic, it triggers the cyclic rules of stress and (vacuous) reduction to form . In Cycle 2, the derivational suffix is spelled out and triggers the cyclic rules of stress and reduction: . But, in Cycle 2, we also spell-out the noncyclic inflectional suffix -ov. Halle and Vergnaud (1987a,b) argue that non-cyclic affixes can’t trigger cyclic rules. They also can’t trigger non-cyclic rules. Thus, no rules apply at all. Finally, all the words undergo a final noncyclic cycle where only noncyclic rules apply. This gets us final stress in .

For compounding, I assume that the compound’s category is not due to adjunction (Selkirk 1982). When combined with Morphological Merger or any other rebracketing process, Halle and Vergnaud (1987a)’s model can handle the bracketing paradox. It can also arguably trigger the right cyclic rules. The problem is that they rely on the Strict Cyclicity Condition (Mascaró 1976), a debunked principle (Kiparsky 1993).

• (59)

  

The exocentric plural has a straightforward derivation. In Cycle 1, the two stems undergo the cyclic rules and get stressed. In Cycle 2, they are concatenated and undergo the cyclic rules of stress and reduction. Note that Rightmost Stress Assignment does not remove primary stress from Stem1. Removal is done by Destressing. In Cycle 3, the plural suffix is added. Because the suffix is non-cyclic, it does not trigger cyclic stress shift and reduction. The word eventually undergoes the non-cyclic stratum and gets final stress.

For the endocentric plural, the first two cycles are straightforward. In Cycle 1, the stems are stressed. In Cycle 2, the second stem gets the plural suffix -er. Because the suffix is non-cyclic, then no rules apply. In Cycle 3, the stems are concatenated. The suffix does not get any stress because of the Strict Cycle Condition (SCC). SCC blocks rightmost stress assignment on the suffix -er because the substring er was created in the previous cycle. We still get destressing and reduction on Stem1’s u because concatenation made u no longer a final vowel. In the final noncyclic cycle, the compound undergoes the noncyclic rules to get final stress.

As shown, Halle and Vergnaud (1987a)’s system can work for Armenian. However, it has three conceptual problems. First, the derivation mainly worked because of the SCC. In Cycle 3, the input was and the output was with reduction on Stem1. In this cycle, the SCC blocked stress shift to the suffix -er, which itself blocked destressing and reduction in *. Current incarnations of Halle and Vergnaud (1987a)’s system still assume that the SCC plays some role (Halle and Nevins 2009). However, the SCC is a dubious principle and has been debunked (Kiparsky 1993). An alternative blocking systems like the Phrase Impenetrability Condition are likewise dubious (Embick 2014; Newell 2017). ²⁹ See Scheer (2011) and Bermúdez-Otero (2008) for a historical overview on the development and failures of the SCC.

Second, Halle and Vergnaud (1987a) assumes that it is arbitrary and unpredictable whether an affix is cyclic or not. This misses the fact that in Armenian all cyclic and non-cyclic affixes are respectively derivational and inflectional morphology. Third, Halle and Vergnaud (1987a)’s system allows cyclic and noncyclic affixes to linearly precede and follow each other, e.g., English patent-abl-ity where -able is Level 2, while -ity is Level 1 (Halle and Kenstowicz 1991; also in Salishan: Czaykowska-Higgins 1993). It is then surprising that these ordering paradoxes are not found in Armenian. All noncyclic affixes are inflectional, and they follow all other derivational morphology.