A Parallel Approach to Mobile Affixation in Huave

This paper examines the system of “mobile affixation” in Huave (isolate, Mexico), specifically the San Francisco del Mar variety described and analyzed by Kim (2008, 2010, 2015a,b). The central properties of Huave’s mobile affixation system can be exemplified by the behavior of the Completive /t/ (CP) when it attaches to simple bases, as shown in (1). When CP /t/ is affixed to a C(...)V base (1a), it appears after the base (i.e., as a “suffix”). On the other hand, when it is affixed to a V(...)C base (1b), it instead surfaces before the base (as a “prefix”). This mobility is, at least in part, conditioned by phonological factors: namely, the affix lands wherever it can avoid a cluster, or the automatic epenthesis that would repair it (Kim, 2008, 2010).1


Introduction
This paper examines the system of "mobile affixation" in Huave (isolate, Mexico), specifically the San Francisco del Mar variety described and analyzed by Kim (2008Kim ( , 2010Kim ( , 2015a. The central properties of Huave's mobile affixation system can be exemplified by the behavior of the Completive /t/ (CP) when it attaches to simple bases, as shown in (1). When CP /t/ is affixed to a C(...)V base (1a), it appears after the base (i.e., as a "suffix"). On the other hand, when it is affixed to a V(...)C base (1b), it instead surfaces before the base (as a "prefix"). This mobility is, at least in part, conditioned by phonological factors: namely, the affix lands wherever it can avoid a cluster, or the automatic epenthesis that would repair it (Kim, 2008(Kim, , 2010. 1 (1) Completive /t/ mobile affixation [(Kim, 2010:140 This phonological characterization is consistent with the behavior of the two other possible base types, V(...)V and C(...)C. 2 In both of these cases, the structures resulting from prefixation and suffixation would be equivalent with respect to cluster phonotactics: when both edges are vowels (2a), no cluster would arise either way; when both edges are consonants (2b), a cluster, which must be resolved via epenthesis, would arise either way. Since the relevant phonological factors do not come into play in these types, they reveal that suffixation is the "default" behavior for these affixes.
(3) a. *CC: Assign one violation * for each sequence of two adjacent consonants. b. DEP: Assign one violation * for each inserted segment. Kim (2008Kim ( , 2010 shows that this can be implemented in a cyclic cophonology approach. This approach specifies a sequence of attachment for all morphemes, which are divided up into "Layers", as schematized in (4). (Some layers require further internal structure.) The Layers marked with dotted lines contain mobile affixes. In this approach, mobile affixes are morphemes whose exponent is a single consonant, and whose cophonology is the one in (5a), where the anti-cluster constraints outrank the relevant alignment constraint. Non-mobile affixes are either those with the cophonology in (5b), where alignment outranks the anti-cluster constraints, or those which consist of just a vowel, and hence do not conflict with anti-cluster constraints.
uc-it *! c. t-uc ** (u,c) CP-eat 's/he ate' (intrans.) [Kim 2010:140, ex. 12b] While no base (root + theme vowel) both begins and ends in a consonant (C(...)C), this configuration does occur when multiple affixes are present (see §2.4). In these cases, schematized in (11), the *CC problem arises regardless of attachment site. The alignment analysis (whether cyclic or parallel) rightly predicts that the affix will surface to the right in these cases. Prefix (11c/d) and suffix candidates (11a/b) respectively tie on either *CC or DEP, and the tie is broken by lower-ranked ALIGN-ASP-R in favor of the suffix (11b). From these examples, we see that epenthesis is preferred to creating a cluster: *CC ≫ DEP (12).  (13), are marked by two Layer 1 morphemes: a fusional 2nd person intransitive marker /r/ (abbrev. "2I"), and a general 2nd person agreement marker /e/ (abbrev. "2"). 4 2 /e/ always surfaces to the left of the base, but the position of 2I /r/ varies by base shape: with C(...)V bases (13a), 2I /r/ surfaces at the right edge; but with V(...)C bases (13b), 2I /r/ surfaces between 2 /e/ and the base. That is to say, 2 /e/ doesn't show mobility (always a prefix), 5 but 2I /r/ does. This is exactly what we expect if mobility is driven by phonotactic pressure to avoid clusters: a consonantal affix moves, because it would create a cluster in its otherwise preferred position; but a vocalic affix always stays in its preferred location, because it will never create a cluster.
(13) Mobility of the 2nd person intransitive in the atemporal paradigm [(Kim, 2008: This behavior can be modeled using multiple alignment constraints, in conjunction with *CC and DEP. 2 /e/ is unwaveringly prefixal, so it must have a left-alignment constraint: ALIGN-2-L. While alternatives exist at the moment, 2I /r/ does have a right-alignment constraint: ALIGN-2I-R. These two constraints, regardless of ranking, derive the C(...)V case (14). Candidate (14c) e-mo h ko-r perfectly left-aligns 2 /e/ and perfectly right-aligns 2I /r/. Because the base ends in a vowel, right-alignment of 2I /r/ does not raise any phonotactic problems. All other candidates fare worse on alignment, and many have phonotactic problems too.
(15) Ranking: c. e-uc-(i)r *! d. r-uc-e *!** (r, u,c) *** (u,c, e) e. e-r-uc ** (u,c) f. r-e-uc *! (r) *** (e, u,c) 2-2I-eat 'you (sg.) eat' (intrans.) [(Kim, 2010:140, ex As already established, both Aspect and 2I /r/ have right-alignment constraints. Likewise, both are single consonants, and respond to the same phonotactic conditions. Since only one morpheme can attain wordfinal position (assuming no coalescence), their relative left/right order tells us the ranking of their alignment constraints. In optimal candidate (20c) t-e-mo h ko-r, 2I /r/ (successfully) surfaces in word-final position, whereas CP /t/ is displaced all the way to word-initial position. This requires the ranking in (19). e-mo h ko-t-or *! ** (o,r) f. e-mo h ko-t-r *! * (r) In most languages, the result of this competition would be a form like *e-mo h ko-t(o)r (20e/f), where both affixes end up as suffixes, in the order determined by the alignment ranking. This would violate either DEP (20e) or *CC (20f), but ordering considerations would prevail. What's different in Huave is the tolerance of mobility in the service of phonological problems. Because the morphology happens to furnish an additional vowel (the 2 morpheme /e/), the cluster problems faced by (20e,f) can be avoided by "movement" of one of the affixes: either CP /t/ could move left, before 2 /e/ (20c); or 2 /e/ could move right, between the mobile affixes (20a). If ALIGN-2-L ≫ ALIGN-ASP-R (which follows from transitivity; cf. (15) and (19)), leftward displacement of CP /t/ (20c) will be preferred. Note, nonetheless, that ALIGN-2-L satisfaction is not perfect. The 2 /e/ morpheme is still displaced from the left edge by CP /t/, just minimally so. This violation is driven by the higher ranking of {*CC ≫ DEP}. The resulting order is determined by optimizing alignment over whatever candidates remain. This leaves us with the ranking in (21). 2nd person intransitive completives of V(...)C bases, e.g. t-e-r-uc (22), follow from the same ranking. Here, the two available vowels are both on the left side of the base, and therefore the two consonantal affixes surface on that side. Nevertheless, the relative linear order of the two consonantal affixes is maintained: 2I /r/ is further right than CP /t/, because ALIGN-2I-R dominates ALIGN-ASP-R. ii.

Mobile 1st person /s/ and its interaction with "Aspect"
The last mobile affix is the 1st person agreement morpheme /s/ (abbrev. "1"). Kim assigns it to Layer 3, because it tends to occur "outside" of the Layer 1 and Layer 2 affixes. (In fact, it is the only Layer 3 affix.) The Layer 2 affix relevant here is the prefixal Future marker /i/. Huave's morphology prevents precise minimal pairs, but the mobility of 1 /s/ can be seen in the contrast between the prefixal realizations in V(...)C bases (23)  . One affix will attach to that vowel (avoiding a cluster), but the other will be stuck next to a consonant. This is the first (real) case where epenthesis is unavoidable. Nevertheless, the relative left/right order of 1 /s/ and Asp reveals the relative ranking of their alignment constraints: 1 /s/ surfaces at the right edge while Asp is displaced all the way to the left; hence (27). Notably, even though Asp is attaching in the phonotactically preferred location, it is 1 /s/ whose alignment preferences are being satisfied. Therefore, under this analysis, there is no direct connection between morphological priority and phonotactic optimization. The two considerations co-exist, but are distinct. s-uk w al-n *! ***** (u,k w ,a,l, n) b. s-uk w al-an * *!***** (u,k w ,a,l, a,n) c.
sa-n-uk w al * *!***** (an, u,k w ,a,l) **** (u,k w ,a,l) l. na-s-uk w al * *!*** (u,k w ,a,l) ****** (a,s, u,k w ,a,l) Crucial rankings, and supporting candidate comparisons, are identified in (30): The morphology again conspires to make a perfect C(...)V base minimal pair unavailable, because the intransitive suffix /u/ is required with such roots. 6 The vowel of ITR /u/ allows for optimal alignment without phonotactic problems. So, ALIGN-1-R ≫ ALIGN-ASP-R (ranking (30c)) determines the outcome on its own (31). This shows that the interactive mobility of 1 /s/ and Completive/Stative can be explained in exactly the same way as the other types of mobility discussed thus far. ii.
Ùutu-n-i-s **!*** (Ù,u,t,u, n) ** (i, s) c. s-i-Ùutu-n * (s) ****** (i, Ù,u,t,u, n) n-i-Ùutu-s * (n) *!**** (i, Ù,u,t,u) The fragment {*CC ≫ DEP} must dominate ALIGN-FUT-L in order to avoid better-aligned epenthesis candidates, e.g. * [i-Ùutu-s-un] for tableau (38). The full rankings (all crucial) are pulled out in (39). The conflation of Completive/Stative with Subordinate is no longer tenable, as evident from the paradoxical rankings in (40). This is because it is purely left/right order that reflects alignment rankings (see Trommer 2001), unlike Kim's cyclic layered system, where sequential attachment is reflected by inside-out ordering. Separating out Subordinate from Completive and Stative is probably desirable. 7 Completive and Stative are clearly aspectual morphemes. But Subordinate explicitly does not carry aspectual information (Kim, 2008:263), and should probably rather be characterized as some sort of mood morpheme (ibid.: Ch. 6.2.1).  researchers working within Base-Derivative Correspondence Theory agree that, in order for a form to function as a base of correspondence, it (at least proto-typically) must (i) be a "free-standing output form" and (ii) constitute "a subset of the grammatical features of the derived form" (Kager, 1999:282). As long as we assume that Huave singulars are underspecified for number (59b), the singular will meet the standard criteria for basehood if we adopt the morphosyntactic structure for the plurals predicted by the MAP (59a).

Conclusion
This paper has demonstrated that mobile affixation in Huave can be modeled with a fully parallel phonological approach, in addition to the "cyclic, layered" approach developed by Kim (2008Kim ( , 2010Kim ( , 2015a. Violable, gradient alignment constraints indexed to individual morphemes, comprising a single language-wide ranking, govern competition between those morphemes for position in the output. 13 This is possible because of a previously unrecognized generalization about the left/right stability of affix order across the language: (60) Huave affix ordering generalization: For any two affixes of the same phonological shape, those affixes' relative left/right order is the same across all forms in which they co-occur.
The "desired" order (in terms of alignment) can be interrupted by the force of higher-ranked constraints against consonant clusters (*CC) and epenthesis (DEP). This explains why only consonantal affixes are mobile. "Mobility" can thus be characterized as excess alignment violation to avoid a cluster/epenthesis. One noteworthy difference from Kim's analysis is that the immobility of the plural morphemes correlates with their order: the ranking of alignment constraints with respect to DEP governs mobility, but it also serves as a pivot for the relative ranking of alignment constraints with respect to each other, which is what determines left/right order. This generalization can only be captured in a parallelist approach. When further tied in with the Mirror Alignment Principle (Zukoff, to appear), this same distinction comports with cross-linguistic tendencies/universals about the relative structure of Person and Number, and considerations of "base-hood" with respect to Base-Derivative Correspondence Theory.