In support of an OT-DM model

Abstract

This paper provides support for a modified DM model which I call Optimality-Theoretic Distributed Morphology (OT-DM). The strongest form of this model is that all morphological operations take place in parallel, which I call the Morphology in Parallel Hypothesis (MPH). Although combining OT and DM is unorthodox in practice, I show that a growing body of data warrants this modification (Trommer 2001a, 2001b, 2002; Dawson 2017; Foley 2017; a.o.). I provide support for OT-DM from the distribution of verbal clitics in Degema, a language of southern Nigeria. Within, I argue that agreement clitics are inserted post-syntactically via the DM operation Dissociated Node Insertion (DNI), and further that verb complexes are formed post-syntactically via the operation Local Dislocation (LD), operating in tandem with a well-formedness markedness constraint which requires verbs to appear in properly inflected words. These DM operations are decomposed into a series of constraints which are crucially ranked. Candidates are freely generated from gen and are subject to all DM operations, and are evaluated via eval against the ranked constraint set. I illustrate that under the standard serial DM model in which DNI proceeds VI, this would result in the wrong output form, and that even after parameterizing DM operation order in response, this model does not adequately capture the motivations behind the morphological patterns.

Appendices

Appendix 1: Evaluation (full tableaux)

[See supplemental files 11049_2019_9444_MOESM1_ESM.pdf, 11049_2019_9444_MOESM2_ESM.xlsx, 11049_2019_9444_MOESM3_ESM.txt, 11049_2019_9444_MOESM4_ESM.txt and 11049_2019_9444_MOESM5_ESM.htm]

Appendix 2: Factorial typology of constraint set

[See supplemental files 11049_2019_9444_MOESM6_ESM.txt, 11049_2019_9444_MOESM7_ESM.xlsx, 11049_2019_9444_MOESM8_ESM.txt, 11049_2019_9444_MOESM9_ESM.txt, 11049_2019_9444_MOESM10_ESM.txt, 11049_2019_9444_MOESM11_ESM.txt, 11049_2019_9444_MOESM12_ESM.txt and 11049_2019_9444_MOESM13_ESM.htm]

Within Optimality Theory, a factorial typology refers to determining all the logical rankings of a set of constraints and computing the different winning sets of output candidates (Kager 1999:35). A factorial typology is particularly useful in determining if there are pathological predictions from a set of constraints. In the current data set, a potential example of a pathological prediction would be a grammar which inserted subject agreement when there is a prosodically light object but not when there is a prosodically heavy one. By default, we expect the presence of subject agreement and the prosodic type of object to be orthogonal. This pathological grammar is not generated by the set of constraints proposed.

A factorial typology was determined using OTSoft v2.5 (Hayes et al. 2013). There were 15 constraints considered, making the logically possible number of grammars the factorial 15! (1,307,674,368,000). Restricting our inputs to the 6 input types presented above, the factorial typology resulted in 128 distinct grammars. The factorial typology does not reveal any straightforward pathological predictions, and therefore this constraint set is sufficiently restrictive. Major parameters of output variation include (1) not incorporating /D_σ/, (2) lacking subject agreement, (3) not incorporating /asp/ but having aspect agreement, (4) not dislocating /asp/ to the right edge, (5) dislocating /V2/ over a pronoun /D/, and (6) not incorporating /V2/ resulting in /V1/ and /V2/ forming separate MWds, with each verb receiving separate inflection i.e. the double-marking pattern.

A grammar of this last type is provided above in Table 12 with Grammar 18 (not the attested Degema patterns). Recall from Sect. 3.2 that the /V V/ and /V D_σ V/ types with double-marking is interpreted as dispreferred ^(?) to ungrammatical * in this dialect of Degema depending on speaker. A grammar which optimizes double-marking here can be derived if we reorder constraint stratum 3 over constraint stratum 2, as shown below (C-Strata order established in the Tableaux in Sect. 5 above).

Table 12 Constraint strata ranking and output candidate which corresponds to $^{(?)\sim ^{?}\ast\sim\ast}$ grammaticality interpretation

A sample of grammars generated by the factorial typology is in Table 13. In Grammar 47 No Asp Inc, /asp/ is not incorporated into a surrounding verb and there is no aspect agreement. In Grammar 15 V Over D, /V2/ incorporates into the MWd containing /V1/ but the object /D/ is not incorporated. In this type, /V2/ actually incorporates over /D/ resulting in a change in word order. On the surface, therefore, such grammars may be ambiguous to speakers whether they involve this type of post-syntactic incorporation, or syntactic head movement like the one discussed above (Sect. 3.3).

Table 13 Sample of grammars from factorial typology

Grammar 74 Inc w/o Agr involves the incorporation of /V2/ into /V1/ without the presence of agreement, i.e. an optimal output ∖(asp) (V+D+V)∖. Such a structure results when constraints requiring verbal inflection (i.e. V=WF_MWd(agr_sbj) and V=WF_MWd(asp)) are ranked below constraints disallowing these morphemes (i.e. *agr_sbj and *agr_asp) but above a mapping constraint requiring terminal nodes map to MWds (i.e. Map(Wd_Type). In this case, incorporation takes place because it has less violations of the V=WF constraints.

Grammar 68 Cond Inc D involves incorporating /D/ only under the conditions that /V2/ is also incorporated, i.e. a non-SVC output ∖(asp) (V) (D_σσ)∖ but a SVC output ∖(asp) (V+D_σσ+V)∖. In this case, /D/ is incorporated only if it would otherwise ‘get in the way’ of V2 incorporating, and not otherwise. The constraint ranking which accounts for this involves ranking the V=WF_MWd constraints and Map(Wd_Type) crucially above MWd=PrWd.

The final two grammars involve morphological labeling, as discussed in Sect. 5.4, and require special attention. In grammar 7 D Label Repair, the optimal output ∖(V+D+V+asp)_{D}∖ bears a label {D} rather than the label {V} in order to avoid a violation for not bearing agreement. This output is found in grammars 7, 14, 16, 27, 29, 36, 43, 45, 50, 57, 59, and 64. Similarly, grammar 70 Multiple Labels involves effects when one of the candidates has multiple labels. In this case when the optimal output candidate has a {V} label, the /asp/ marker surfaces appears in its own MWd, whereas in another derivation the optimal output has {V}, {D}, and {Asp} labels and /asp/ appears at the right edge of this MWd. This output is found in grammars 39, 46, 53, 60, 70, 77, 84, and 91.

It is unclear at this point whether the last two types of grammars constitute pathological predictions of the constraint set. This is because although candidates were generated systematically along several important dimensions (described in Sect. 5.2), generating the complete list of candidates with different morphological labels was not feasible for this project. I therefore suspect that these final two grammar types are merely an artefact of the limited number of candidates evaluated in this study with respect to different morphological labels.

Appendix 3: Against two alternatives to the OT-DM analysis

3.1 A3.1 Alternative 1: Syntactic verb movement

One alternative to the OT-DM analysis is what I refer to as the syntactic verb movement alternative. Recall from the discussion of the syntax of Degema SVCs in Sect. 3.3 that I assume a vP complementation structure of SVCs where V₁ selects v₂P as its complement: [_aspPasp⁰ [_vPv₁⁰ [ V₁⁰ [_vPv₂⁰ [ V₂⁰ ] ] ] ] ] (Collins 1997, 2002). I have called verbs in the single-marking clitic pattern a morphological compound based on their acting as a single constituent with respect to clitic marking (and grammatical tone, see Sect. 4.4). Under my analysis, both single-marking and double-marking SVCs share the same syntactic structure and are subject to the same sequence of syntactic operations. In other words, both patterns are syntax-equivalent up to the point of spellout.

In contrast, the syntactic verb movement alternative derives the constituency of the two verbs by overt syntactic head-movement, whether by V2 moving to V1 directly, or both V1 and V2 moving to the same higher functional head such as v1. Under this alternative, single-marking is attributed to the verbs forming a syntactic compound and therefore being inflected like non-compound verbs with a single set of clitics. Double-marking is attributed to the lack of syntactic verb movement under specific conditions resulting in the two verbs being spelled-out as separate words.

Syntactic verb movement in SVCs is argued for by Collins (2002) in his analysis of ǂHoan verb compounds [huc] (Kxa: Botswana), exemplified in (41). Here, the verbs cluster in a pre-object field and are marked by a single aspect marker a−. Both arguments of the verbs appear afterwards.

• (41)

  

I refer the reader to the original paper for syntactic details of this analysis.

This alternative is attractive in that it nullifies the need for local dislocation sensitive to particular lexical categories, and unifies the application of dissociated node insertion of agreement heads without stipulation. This syntactic alternative therefore provides us the best chance of success in accounting for the Degema patterns by appealing only to syntax, without the need for post-syntactic constraints, optimization, and output-sensitivity.

3.1.1 A3.1.1 Argument 1: Phonologically null objects show single-marking pattern

The first argument against the syntactic alternative is that phonologically null objects of V1 show the single-marking pattern. I demonstrated above that there is no semantic difference between the single- and double-marking pattern: all SVC types may show either pattern depending on the surface order of nouns and verbs, e.g. examples (6)–(7) above. Consider the data in (42) involving question words in SVCs. When the question word ovo ‘who’ appears in-situ in object position (ex. (a)), the verbs are not sufficiently local and therefore cannot form a single MWd. As a consequence, each is marked with a set of clitics, the double-marking pattern. When the question word appears ex-situ in a cleft construction, no object then intervenes between the verbs and they become sufficiently local and appear with single-marking (ex. (b)). It is ungrammatical in this ex-situ context to mark the verbs with double-marking.

• (42)

  

Likewise, example (43) shows that single-marking surfaces when an object is not present after V1 due to focus clefting (a), relativization (b), or object pro-drop (c). This last example is especially telling as it involves object drop from two transitive verbs, resulting in five sufficiently local verbs. All five verbs form a MWd, with one proclitic on V1 and one enclitic on V5, and a H tone melody over the entire verbal complex.

• (43)

  

These data illustrate that clitic marking is insensitive to the selectional properties of the verbs, e.g. transitive vs. intransitive verb roots. If single-marking were the result of syntactic head movement, movement of the lower V2 head upwards would be triggered by a feature of a higher functional head. Under standard Minimalist assumptions, this predicts that by default when the syntactic structural condition is met, verb movement takes place. We therefore expect for single and double-marking patterns to be stable in the absence of an intervening object, as the presence of an object is orthogonal to the presence of a strong feature on a functional head. This is in particular expected under a Copy Theory of Movement (Nunes 1995), where ‘traces’ are simply lower copies of moved constituents and are present in the syntax, but deleted at spellout. These expectations would not be borne out under the syntax-only alternative.

3.1.2 A3.1.2 Argument 2: Unmotivated ‘blocking’ of head movement by an overt object

The second argument is that there is unmotivated ‘blocking’ of head movement by an overt object between V1 and V2, a counterpart to the first argument. Under the syntactic head movement alternative, V2 undergoes movement when triggered by a strong feature on a higher functional head. This movement therefore should be insensitive to whether there is an overt object present. Recall from the ǂHoan verb compound example in (41) above that both verbs are subject to movement past their arguments, resulting in a [V1+V2 O1 O2] order (whether in complement or specifier position). In general, in languages exhibiting verb compounding which (superficially) resemble the ǂHoan type, the presence of an overt object does not block verb compounding (see Rolle and Kari 2016:155 for languages/references).

We can compare this to the Degema facts. In Degema, V2 does not move over an intervening object to form a constituent with V1. For example, in (44) the transitive verbs tạm ‘chew’ and ḍọny ‘swallow’ appear in a SVC and share the internal object ịḍíyōm ‘food’. The object appears in its expected position between the verbs. It is ungrammatical to move V2 past the object, as in b. Similarly, (45) shows that verb movement is equally disallowed with a transitive V1P and an intransitive V2. Here, V1 and V2 cannot appear adjacent.

• (44)

  

• (45)

  

It is unclear how the presence of an object in this specifier position could act as a syntactic blocker for head-movement, and weakens support for the syntactic movement alternative.¹⁷

3.1.3 A3.1.3 Argument 3: SVCs with prosodically light pronouns show single-marking

A third argument against V2 head movement involves prosodically light pronouns. Recall that although objects generally block the single-marking SVC pattern, intervening prosodically light pronouns require it, i.e. the cl=[V₁ D_σ V₂]=cl pattern. Under the syntactic movement alternative, because V1 and V2 form a single syntactic word via head-movement, and because D_σ still intervenes between them, it would have to be the case that D_σ also undergoes syntactic movement to this same complex head, a type of syntactic pronoun incorporation.

This is problematic for a number of reasons. First, recall that only prosodically light pronouns show incorporation; prosodically heavy pronouns condition the double-marking pattern. If under the syntactic movement analysis pronouns are also subject to syntactic incorporation, it is not clear how the syntactic trigger could only target prosodically light pronouns rather than all pronouns with an appropriate feature [D]. Under DM assumptions, syntax would not have access to phonological information before spellout. Moreover, recall that the light pronouns do not form any natural class either with respect to their morphosyntactic features. Light pronouns expone feature bundles 1sg, 2sg, 2pl, and 3pl, while heavy pronouns expone 3sg and 1pl (see Table 3).

Further, monosyllabic pronouns and bisyllabic pronouns do not exhibit different syntactic behavior. For example, prosodically light pronouns do not behave like syntactically incorporated pronouns, e.g. they are able to undergo clefting to express focus.

• (46)

  

If the pronoun internally merges with a verb via head movement, and then subsequently moves out of that complex, this would be a type of ‘excorporation’ (Roberts 1991, 2011). There is very limited evidence for bona fide excorporation in the literature, with many arguing against the possibility of it (e.g. Julien 2002:67–87 and references therein; Matushansky 2006:95; a.o.).

3.2 A3.2 Alternative 2: Deletion-under-identity of intermediate clitics

Another alternative is what I term the deletion-under-identity (DUI) alternative. Under this alternative, the grammar uniformly generates a full set of proclitics and enclitics on all verbs in a SVC, but under specific conditions involving identity there is obligatory deletion of intermediate clitics (assumed to be a type of ellipsis). This is schematized in (47) below.

• (47)

  

DUI would take place under two conditions: (1) the deleted clitics appear adjacent, and (2) the clitics be featurally identical. One advantage of this alternative is that Dissociated Node Insertion would take place uniformly on all verbs, and avoid any need to license Local Dislocation of verbs. This would therefore be compatible with a strictly serial rule-based analysis, and not warrant modifying core DM architecture. Such an alternative is flexible and could come in two flavors. The first would be that each verb’s extended projection in the SVC contains a separate Asp⁰ head; therefore the second aspect morpheme would not be agr_asp (see discussion in Sect. 3.3). The second would be that there is one Asp⁰ head, followed by concord aspect agreement. Both of these structures would result in the same input to a DUI operation.

There is precedence for similar types of DUI, e.g. coordination/conjunction reduction (Merchant 2012), ‘suspended affixation’ (Kabak 2007; Guseva and Weisser 2018), and certain cases of verbal ‘unbalanced coordination’ (Johannessen 1998). An example of coordinate reduction in German nouns is in (48) below, where the medial morpheme schaft adjacent to the coordinator is deleted under identity.

• (48)

  

I present two arguments against a DUI alternative. The first is that this analysis makes the wrong predications and overgenerates with respect to a fuller set of Degema data. For example, one context which meets the surface conditions (adjacency and featural identity) is covert coordination and other conjoined clauses. Covert coordination involves two clauses adjacent without a phonologically overt coordinator as in (49)a. The second clause can involve a linker auxiliary, e.g. kị́rí ‘also.aux’ in (49)b. However, in these contexts DUI is not found and is in fact ungrammatical.

• (49)

  

The following example in (50) comes from a Degema text (Kari 1997). In this context, two events are being described—one moving and one settling—each expressed in Degema by two SVCs in two clauses. Within each SVC, the verbs appear adjacent and therefore show single-marking. However, across the SVCs, because they appear in distinct clauses they are not conflated into one larger single-marking structure. Pronunciation of this sentence reveals that there is an obligatory pause between the final verb of SVC1 ḍẹsi ‘go far’ and the first verb of SVC2 tạ ‘go’ (E.E. Kari, p.c.), suggesting they are distinct constituents.

• (50)

  

Under the DUI alternative, it is paradoxical why ellipsis does not take place across clauses as well as within them (cf. English She_italked to me then Ø_ileft).

Further, under the DUI analysis the Degema single-marking pattern involves simultaneous backward deletion (deletion of material in the first conjunct) and forward deletion (deletion of material in the second conjunct), as discussed by Wilder (1995, 1997). Wilder illustrates that backward deletion and forward deletion are distinct operations subject to different phonological, syntactic, and semantic conditions, and we can therefore understand them as independent operations. However, in Degema when clitic deletion takes place, backward and forward deletion must take place simultaneously, as they do not occur without each other. For example, in example (51) below, the proclitics share featural identity and the second proclitic appears at a conjunct boundary, identified as a common condition for DUI processes. However, deletion of the second proclitic is ungrammatical here and shows no variation across speakers. The DUI alternative therefore overgenerates.¹⁸

• (51)